See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266354681

Mesoscopic Evacuation Modeling for Small- to Medium-Sized Metropolitan

Areas

Article

CITATIONS READS

19 942

2 authors, including:

Mohammad Naser
University of Jordan
12 PUBLICATIONS   95 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Mohammad Naser on 22 May 2016.

The user has requested enhancement of the downloaded file.

 Mesoscopic Evacuation Modeling for Small-
to Medium-Sized Metropolitan Areas

Mohammad Naser, Ph.D.*

Shawn C. Birst, PE*

Advanced Traffic Analysis Center

Upper Great Plains Transportation Institute
North Dakota State University
Fargo, ND 58108

August 2010

*At the time of publication, Mohammad Naser is an Assistant Professor with the University of Jordan (Amman) and
Shawn Birst is a Senior Transportation Engineer with Reynolds, Smith and Hills, Inc.
Disclaimer

The contents of this report reflect the views of the authors, who are responsible for the facts and the
accuracy of the information presented herein. This document is disseminated under the sponsorship of the
Department of Transportation, University Transportation Centers Program, in the interest of information
exchange. The U.S. Government assumes no liability for the contents or use thereof.

North Dakota State University does not discriminate on the basis of race, color, national origin, religion, sex, disability, age, Vietnam Era Veteran's status, sexual
orientation, marital status, or public assistance status. Direct inquiries to the Vice President of Equity, Diversity, and Global Outreach, 205 Old Main, (701) 231-7708.
ABSTRACT
Modeling and developing different evacuation scenarios gained significant interest in recent years, where
the management of the transportation system in support of evacuation efforts proved to be critical in
mitigating the impacts of regional emergency events. However, most of the evacuation modeling efforts
focused on large urban areas due to the recurrence of the regional emergency event, resources availability,
and data availability. For urban areas that are classified as small- and medium-size metropolitan areas,
confusion still exists at the metropolitan planning organization (MPO) level on how to develop effective
and practical evacuation plans.

This study aims to develop a methodology for supporting effective decision making and testing
emergency scenarios while taking into account various factors and their effect on public safety. The focus
of this study is on developing an evacuation model for urban areas utilizing the resources available to
MPOs and obtaining local evacuation data, which include human behavior data from a local household
survey.

A case study is developed using Fargo-Moorhead Council of Government’s travel demand model
integrated with DYNASMART-P software. The modeling approach provides direct connectivity with the
regional model, and the hybrid model incorporates a traffic generation component into the regional model
along with dynamic supply.
TABLE OF CONTENTS
1. INTRODUCTION................................................................................................................................... 1
1.1 Problem Statement............................................................................................................................. 2
1.2 Study Objectives ................................................................................................................................ 3
1.3 Report Organization .......................................................................................................................... 3

2. LITERATURE REVIEW ...................................................................................................................... 5

2.1 Fundamentals of Evacuation Planning .............................................................................................. 5
2.1.1 Urban travel demand modeling and evacuation modeling ......................................................... 5
2.1.2 Evacuation modeling parameters ............................................................................................... 8
2.1.3 Modeling tools for evacuation planning..................................................................................... 9
2.1.4 Human factors and evacuation modeling ................................................................................. 12
2.1.5 Travel surveys and evacuation modeling ................................................................................. 13
2.2. Evacuation Modeling....................................................................................................................... 15
2.2.1 Evacuation travel demand ........................................................................................................ 15
2.2.2 Evacuation trip loading rates .................................................................................................... 16
2.2.3 Selection of evacuation modes and routes ............................................................................... 18
2.3 Description of Software Tools......................................................................................................... 18
2.3.1 DYNASMART-P software ...................................................................................................... 18
2.3.2 Cube voyager software ............................................................................................................. 23

3. METHODOLOGY AND DATA ANALYSIS .................................................................................... 25

3.1 Methodology.................................................................................................................................... 25
3.1.1 Modeling approach................................................................................................................... 25
3.1.2 Model development .................................................................................................................. 26
3.1.3 Daily travel demand data.......................................................................................................... 27
3.1.4 Regional travel demand model ................................................................................................. 27
3.1.5 Regional travel demand model calibration............................................................................... 31
3.1.6 Development of the mesoscopic simulation model .................................................................. 33
3.1.7 Development and calibration of the hybrid model ................................................................... 34
3.1.8 Emergency evacuation scenario modeling ............................................................................... 36
3.2 Human Behavior Data Collection.................................................................................................... 38
3.2.1 Data collection.......................................................................................................................... 38
3.2.2 Survey population and sampling .............................................................................................. 38
3.2.3 Survey design and administration ............................................................................................ 38
3.2.4 Data analysis ............................................................................................................................ 39

4. CASE STUDY AND MODELED SCENARIOS ................................................................................ 47

4.1 Case Study ....................................................................................................................................... 47
4.2 Fargo-Moorhead Regional Travel Demand Model ......................................................................... 48
4.2.1 Trip generation ......................................................................................................................... 49
4.2.2 Trip distribution........................................................................................................................ 51
4.2.3 Mode split and temporal distribution ....................................................................................... 51
4.2.4 Traffic assignment .................................................................................................................... 52
4.2.5 F-M COG regional travel demand model calibration............................................................... 52
4.3 Fargo-Moorhead Hybrid Model ...................................................................................................... 55
4.3.1 Fargo-Moorhead hybrid model calibration .............................................................................. 56
 4.4 Evacuation Scenarios....................................................................................................................... 58
4.4.1 FARGODOME evacuation scenario development .................................................................. 58
4.4.2 Emergency evacuation scenario development ......................................................................... 59
4.4.3 Emergency evacuation scenario modeling ............................................................................... 61

5. CONCLUSION AND SUMMARY OF THE RESULTS .................................................................. 67

5.1 Summary of Model Development ................................................................................................... 67
5.2 Research Contributions ................................................................................................................... 68
5.3 Recommendations for Future Research ........................................................................................... 69

REFERENCES .......................................................................................................................................... 71

APPENDIX A: Fargo-Moorhead Emergency Evacuation Survey ...................................................... 75

APPENDIX B: Fargo-Moorhead Emergency Evacuation Survey Results ......................................... 81

LIST OF TABLES
Table 3.1. Modeled capacities for urban and rural roads ............................................................................ 30
Table 3.2. Households Compliance with Evacuation Orders ..................................................................... 40
Table 3.3. Compliance Rates by Household Category ............................................................................... 40
Table 3.4. Number of Vehicles Used per Household for Evacuation ......................................................... 40
Table 3.5. Number of Vehicles Used for Evacuation by Household Category .......................................... 40
Table 3.6. Chi-Square Goodness-of-Fit Test for Categorical Variable....................................................... 41
Table 3.7. Chi-Square Test for Number of Cars Used vs. Household Type ............................................... 42
Table 3.8. Households Choice of Evacuation Destination (Modified) ....................................................... 43
Table 3.9. Chi-Square Goodness-of-Fit Test Destination Choice............................................................... 44
Table 3.10. Percentages Evacuating to Each Destination by Household Category .................................... 45
Table 3.11. Time Needed to Prepare of Evacuation ................................................................................... 45
Table 4.1. Vehicle Trip Generation Rates................................................................................................... 49
Table 4.2. Trip Attraction Rates.................................................................................................................. 49
Table 4.3. University Trips Generation ...................................................................................................... 50
Table 4.4. Total Adjusted Productions and Attractions by Trip Purpose ................................................... 51
Table 4.5. Peak Hour Percentage of Daily Traffic ...................................................................................... 52
Table 4.6. Friction Factors Model Coefficients for the F-M COG Model .................................................. 53
Table 4.7. VMT Values by Jurisdiction for the F-M COG Model ............................................................. 54
Table 4.8. Screenline Counts and K Factors for the F-M COG Model....................................................... 55
Table 4.9. Traffic Volume Comparison for the F-M COG Model.............................................................. 55
Table 4.10. RMSE Value by Volume Range for the F-M COG Model...................................................... 55
Table 4.11. Model Assignment by Traffic Volume Range ......................................................................... 58
Table 4.12. Traffic Network Changes at Different Flood Levels ............................................................... 61
Table 4.13. Minnesota Side Evacuation Results ......................................................................................... 62
Table 4.14. North Dakota Side Evacuation Results .................................................................................... 62
Table 4.15. North Dakota and Minnesota Side Evacuation Results ........................................................... 62
Table 4.16. Evacuating Traffic Volumes at Key Locations for the Different Scenarios ............................ 63
Table 4.17. North Dakota and Minnesota Side Evacuation Results (Modified Control)............................ 64
Table 4.18. North Dakota and Minnesota Side Evacuation Results (National Rates) ................................ 65
LIST OF FIGURES
Figure 2.1. Evacuation departure times curves (28). .................................................................................. 17
Figure 2.2. DYNASMART-P assignment procedure (20) .......................................................................... 20
Figure 2.3. Type 1 modified Greenshields' model (20). ............................................................................. 22
Figure 3.1. The modeled area traffic analysis zones ................................................................................... 28
Figure 3.2. The four-step travel demand model .......................................................................................... 29
Figure 3.3. Regional travel demand model calibration. .............................................................................. 32
Figure 3.4. The structure of the hybrid model. ........................................................................................... 35
Figure 3.5. Evacuation modeling methodology. ......................................................................................... 37
Figure 3.6. Evacuation destinations for a flood event................................................................................. 43
Figure 3.7. Mobilization time curves. ......................................................................................................... 46
Figure 4.1. The modeled Fargo-Moorhead metropolitan area. ................................................................... 48
Figure 4.2. Friction factor functions used for the F-M COG model. .......................................................... 53
Figure 4.3. Screenshot from Fargo-Moorhead DYNASMART-P model. .................................................. 56
Figure 4.4. Intersection locations used to collect traffic volume data (Fargo, ND). ................................... 57
Figure 4.5. FARGODOME evacuation DYNASMART-P links/nodes..................................................... 59
Figure 4.6. The 100-year flood level in the Fargo-Moorhead area (38.2 ft). .............................................. 60
Figure 4.7. Location of modified traffic signal timing plans ..................................................................... 64
1. INTRODUCTION
Evacuation is among the major protective actions used in cases of regional emergencies (28) and it is
defined as “the withdrawal action of persons from a specific area because of real or anticipated threat or
hazard” (23). Modeling and developing different evacuation scenarios have gained significant interest in
recent years, after the terrorist attacks of September 11, 2001, hurricanes Rita and Katrina, and the
California wild fires (33). The management of the transportation system in support of evacuation efforts
proved to be critical in mitigating the impacts of these events.

An essential element of any evacuation plan is a carefully prepared transportation plan, where accurate
preplanning adds to the effectiveness of the evacuation process. Evacuations in cases of regional
emergencies result in drastic shifts in the demand for travel over the transportation system. Therefore, the
evacuation travel times are greatly affected by the available capacity of the roadway links. The plan needs
to address both the response and recovery stages, deal with hazards, and communicate the information to
the public.

The main objectives of evacuation planning are to identify the best evacuation routes, and provide
estimates of the time needed to evacuate the at-risk population. Decision makers need to know the
evacuation time estimates (ETEs) associated with different scenarios to provide sufficient warning to the
population and reduce their vulnerability to the hazard. This knowledge helps them to better assess the
danger and reach a balance between the last possible time before they issue the evacuation orders to avoid
costly unnecessary evacuation while having enough time to evacuate without risking loss of life (14).

An evacuation plan must conform to the criteria set by federal, state, and local agencies with jurisdictional
authority and require the coordination among these agencies. There is a great need to establish
coordination and communication among different agencies to promote security and safety while enabling
the transportation system to deal with the public demand for travel. Congress recognized the importance
of this coordination and legislated requirements for transportation planning agencies to address security
issues in their plans, including evacuation planning (31). Metropolitan planning organizations (MPOs)
have an advantage when it comes to technical analysis, local transportation system planning, and
coordination of emergency management efforts.

The Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (SAFETEA-
LU) was introduced in 2005 (41). One of the major changes introduced by SAFETEA-LU was the
separation of the “Safety and Security” planning factor into two stand-alone planning factors for MPOs.
Transportation security refers to both personal and homeland security, with the focus on the vulnerability
to intentional attack or natural disasters and associated evacuation procedures that should be incorporated
in the project selection process. Safety refers to reducing the number of crashes and accidental deaths or
injuries associated with the operation of surface modes.

At the MPO level, the inclusion of security in the transportation planning process is in its early stage and
confusion exists regarding how the federal legislation can be implemented (12). Most of the applications
of security planning are on the operational level of the transportation infrastructure with modest
consideration of the security aspect in the development of transportation improvement Plans (TIPs). In a
survey of the MPOs conducted in 2002, 78% of the surveyed MPOs introduced changes to their planning
process because of security concerns. Also, 25% of the MPOs reported an increase in the costs associated
with the planning process due to the increase in technical considerations and the number of stakeholders
involved in the process (12).

1
1.1 Problem Statement
The development of evacuation plans should be done well in advance of the emergency and be revised
during the emergency itself based on the details and actual conditions of the event. Emergency officials
usually report traffic problems during regional emergency evacuations, such as inadequate roadway
signage, uncoordinated traffic lights, and inadequate traffic control (37). The ability to effectively respond
to or recover from an emergency situation is strongly related to how actively the transportation system
can be managed. For example, the region-wide evacuation traffic management plan developed by
department of transportation and emergency management agencies in the state of Louisiana was
considered to be the primary factor contributing to the effectiveness of recurring hurricane evacuation
events (6).

The effectiveness of an evacuation process is measured by the estimate of the time required for
evacuation. This is defined as the time required to evacuate the at-risk population to a farther, safer area
(53, 11). Basic traffic engineering principles are used to develop ETEs and estimate the transportation
system’s ability to accommodate the evacuation demand.

Most of the evacuation modeling efforts were focused on large urban areas (especially in hurricane
regions) due to the recurrence of the regional emergency event, resources availability, and data
availability. For urban areas that are classified as small- and medium-size areas, there is still confusion at
the MPO level on how to develop effective and practical evacuation plans that could be included in their
long range transportation plans. Because federal requirements include security as a factor to be considered
in transportation planning processes at the metropolitan level, those MPOs need to address emergency
relief and disaster preparedness plans, strategies, and policies that support homeland security (54).

Several modeling approaches have been utilized to address the performance of the transportation system
and estimate evacuation times. However, simulation-based models are among the most powerful tools
that could be applied for accurately and realistically capturing driver-network interactions. Nevertheless,
some challenges face the current simulation-based modeling efforts. These challenges include the high
cost associated with obtaining the software, training staff, and updating the models to reflect changes on
the ground. Also, the availability of data needed for evacuation simulation modeling, which is not usually
covered in the transportation census database, continues to be an obstacle. The main focus of the research
related to evacuation modeling was on developing simulation models to estimate the clearance times
associated with different emergency scenarios. In addition, this study aimed to generate the data needed
for evacuation model development which was not available in the literature or the transportation census.

Evacuation models need to address not only the transportation aspects of the problem, but also the human
behavior aspects associated with emergency evacuations. Also, these models require evacuation traffic
demand data as input, but there was little effort made to incorporate travel demand estimation in those
models (50). Many of the developed evacuation simulation models ignored the actual human behavior
aspects on the ground during emergency evacuations. Human behavioral analysis is critical to obtain
effective and accurate evacuation models and develop reliable estimates of the time needed for
evacuation. It is a major factor in identifying the demand for travel and the demand loading rates on the
transportation system during an emergency evacuation scenario, and there is a great need for more human
behavior data that could be used to develop simulation models.

2
1.2 Study Objectives
This study aims to develop a methodology for supporting effective decision making and testing different
emergency scenarios while taking into account the various factors and their effect on public safety. The
focus of this study is on developing an evacuation model for urban areas utilizing the resources available
by MPOs. In addition, this study aims to develop the human behavior data needed for evacuation
modeling and identify the effect of using those data instead of data developed for hurricane regions to
model regional evacuations. Specifically, the main objectives of this study include the following:
1) Identifying the different local data and population parameters needed for estimating evacuation trips
generation, trips distribution, and trips loading rates using local surveys
2) Developing an emergency evacuation model that is capable of predicting traffic conditions during
regional emergency evacuations recognizing MPOs resources and data availability
3) Evaluating the effects of different traffic operations measures on the performance of the
transportation system during emergency evacuation events
4) Evaluating the effects of using the emergency evacuation data generated for medium size MPOs
compared with using national evacuation data in developing evacuation models on the performance of
the evacuation model

The approach may be applied to the different stages of the emergency response and preparedness: disaster
scenario analysis and mitigation preparedness. The output measures of effectiveness (they include the
predicted traffic volumes, delays, average speeds, and ETEs) associated with different traffic management
options for an emergency evacuation scenario will be used to identify the effect of traffic management
options implemented for the emergency scenario.

Evacuation plans will allow for the evacuation of the population under the safest possible conditions
within the least amount of time. Additionally, the methodology developed in this study has the ability to
account for network and traffic dynamics with reasonable data requirements. Also, it provides
connectivity with the four-step travel demand model by integrating it within the regional travel demand
model, reducing the complexity associated with updating and maintaining the evacuation model.

1.3 Report Organization

This report consists of five sections, including this introduction. Section 2 provides a review of the
literature and the fundamentals of evacuation planning and existing evacuation models. Section 3
provides a detailed description of the research approach and a discussion of the methodology and data
analysis. Section 4 details the case study used for this study, in addition to the model development. The
summary of the results and conclusion of this study is provided in Section 5.

3
4
2. LITERATURE REVIEW
This section provides a review of the fundamentals of evacuation planning. It describes the evacuation
modeling process, modeling parameters, simulation tools, and human behavior during emergency
evacuations. Also, it provides a background on the study framework, which includes identifying travel
demand, trip loading rates, and selection of evacuation destinations and modes.

2.1 Fundamentals of Evacuation Planning

Regional emergency events could be categorized according to the source of the threat (natural or man-
made) and the availability of warning time (long or short). Natural disasters include hurricanes, tsunamis,
floods, earthquakes, and fires. Man-made threats include hazardous material spills, malfunctions of
nuclear plants or chemical facilities, and terrorist attacks. Hurricanes and floods are usually associated
with extended warning times before the danger threatens the safety of the population while hazardous
materials (hazmat) spills, terrorist attacks, and earthquakes have very little warning time to the public, if
any. Different transportation related measures are associated with different phases of the emergency
event, which include the following (48):
1) Pre-disaster
a) Data collection
b) Infrastructure assessment
c) Disaster scenario analysis
d) Mitigation preparedness
2) During disaster
a) Disaster assessment
b) Traffic network assessment
c) Traffic management
d) Evacuation preparation
e) Evacuation deployment
3) Post-disaster
a) Infrastructure assessment
b) Post-disaster traffic management

Evacuation planning was originally required for the nuclear power industry following the partial
meltdown of a reactor at the Three-Mile Island nuclear power plant in 1979. After that incident, the
mandate by the Nuclear Regulatory Commission (NRC) for evacuation plans in the areas surrounding
nuclear power plants gave the first momentum to evacuation modeling (2).

2.1.1 Urban travel demand modeling and evacuation modeling

The main purpose for analyzing travel behavior is to design models that reproduce, as accurately as
possible, the travel flows generated by users of the transportation system and changes in their travel
patterns as a result of changes in the conditions of the transportation system. In areas qualifying for an
MPO, existing metropolitan travel demand models enhanced by different traffic simulation tools provide
a good base for developing evacuation plans. Traditional transportation planning focuses on the diurnal
rhythm travel demands of the average weekday to provide acceptable levels of service throughout the day.
Emergency planning is different from conventional planning applications in that it usually involves
moving a large number of people over a large geographic area during a limited period of time.

Evacuation simulation models play an important role in planning, analyzing, and understanding the
different aspects of emergency evacuations. Most of the modeling and simulation tools used for

5
evacuation modeling are based on the four-step travel demand model. It consists of four major stages that
are related to the user’s trip decision-making process. The first two steps are related to the nature of the
land-use patterns, while the last two steps are dependent on the attributes of the modeled transportation
network. The four stages are trip generation, trip distribution, modal split, and traffic assignment.

2.1.1.1 Trip generation models: These models use social and economic data to predict the number of
trips produced by and attracted to each zone within the study area. Trip production is associated with
residential areas and different attributes of households, whereas trip attraction is related to non-residential
area characteristics. Production is estimated using factors such as the number of households in that area,
household sizes, income, and automobile ownership rates, along with other variables. Attraction is
estimated using variables such as employment levels and floor space (29).

To establish the relationship between trip generation rates and the social and economic data of each zone,
trip generation models use historical data to estimate the number of trips generated. The most commonly
used reference for that is the trip generation manual developed by the Institute of Transportation
Engineers (ITE), which provides the type of data used, trip rates, and other related statistical data (29).
However, the data provided in that manual represent national averages, which are not necessarily
applicable to the local area of interest. In addition, the rates provided in the manual offer a wide range of
values, making it difficult to pick a single value representative of local conditions.

For local area studies, two major types of generation models are used: regression models and category
analysis models (29). Regression models involve techniques such as the least-square regression model.
They are used to predict the trip rates for spatial units or traffic analysis zones (TAZs), based on the
available data for one or more variables, such as the number and size of households in each TAZ. These
models are easy to construct and use, but they do not always yield accurate results. Category analysis
models classify the households and employment type based on similar social and economic attributes
instead of spatial zones. Households with similar attributes are assumed to have the same trends for trip
production, and the same applies to employment type for trip attractions.

For evacuation trips, the number of trips produced is based on the number of households, the nature of the
threat, automobile ownership, and the number of drivers in the household. As for evacuation trip
attractions, they are based on the nature of the threat in addition to the type of destination chosen by the
evacuee. Evacuation trip generation modeling started in the 1990s where it typically involved the
conversion of the number of evacuees into the number of evacuation trips (44). Later, logistic regression
and fixed trip rates were used in several studies (36).

2.1.1.2 Trip distribution models: After the trip production and attraction for each zone is determined in
the trip generation step, trip distribution models are used to connect trip ends, that is, to establish the flow
of trips from production zones to attraction zones (29). The trip distribution in travel demand models
between different trip origins and destinations is based on the purpose of the trip. This is because daily
trips are made to participate in the different daily activities where the choice of trip destination is
determined by the purpose of the trip. The output from this step is a matrix representing the production
and attractions between traffic analysis zones, called the origin-destination (O-D) matrix.

The most commonly used type of trip distribution model is called the gravity model (23). This model is a
modified version of Newton’s law of gravitation between physical bodies in space. In it, the number of
trips between zones is assumed to be based on the levels of activity and relative attractiveness of zones,
which is measured by travel time or cost (50). The major criticism of the gravity model is that it is based
on observation, and cannot be confirmed scientifically. Also it is unable to account for the behavioral
assumptions in predicting trip flows for congested links in the transportation network (27).

6
The destinations for the trips evacuating the at-risk area are based on the nature of the threat in addition to
the type of destination chosen by the evacuee. In modeling evacuation trips, there are three major choices
for trip destinations: closest point out of the danger area, destinations pre-specified by the evacuation
plan, and a mix of options based on factors such as network conditions. The available literature suggests
that most of the evacuees will pick the house of a relative or a friend, followed by hotels and motels, and
a small percentage of them will pick public shelters as their evacuation destination (23).

2.1.1.3 Modal split: These models predict the mode of travel that is used for trips. They divide trips
among the various transportation modes available for users. If the modal split models are applied after the
trip distribution step in urban transportation modeling system (UTMS), then they are called trip-
interchange models (29). For the purposes of this study, transit users do not have a significant effect on
the overall performance of the transportation system. Hence, the only mode of transportation modeled for
the purposes of this study is private automobiles.

2.1.1.4 Traffic assignment: In this step, the predicted traffic flows are assigned to the modeled network
links. Traffic assignment follows the main principles of equilibrium stated by Wardrop in the 1950s (42):
User Equilibrium and System Equilibrium. In User Equilibrium (UE), users of the system choose the
route that would minimize their cost (or travel time) without consideration to the overall average travel
time on the system. In System Equilibrium (SE), system users would behave cooperatively in choosing
their own route to ensure the most efficient use of the system, thus optimizing the overall average cost of
travel on the system.

Static assignment does not represent the traffic flows on the transportation network through time, and all
traffic is assigned simultaneously to all network links along their path. This assumption is not realistic,
but for long-term regional planning purposes, static methods could produce results with a satisfying
degree of accuracy. Dynamic assignment techniques introduce the time dimension into the modeling
process and are usually used for short-term applications that require a higher degree of accuracy.

Static traffic assignment determines the traffic loading on the roadway network in a steady state setting,
while dynamic traffic assignment (DTA) deals with the same problem in dynamic setting (13). DTA-
based models witnessed increasing emphasis due to their potential to address longstanding problems with
unrealistic assumptions of static planning methods. This is primarily because DTA-based models depart
from the standard static assignment assumptions to deal with time-varying traffic flows, which include the
effects of roadway congestion. The data requirements for DTA models are similar to static traffic
assignment models; however, DTA models require path-delay based on the time of departure and traffic
conditions on the trip route instead of volume-delay functions used in static assignment (13).

The DTA tools are mainly used for short-term planning applications. These tools are based on a dynamic
traffic assignment algorithm that can handle both traffic dynamics and travel behavior. In addition, they
provide an effective platform for evaluating different operational strategies and analyzing their
performance and benefits. There are two major methodological approaches provided in the literature for
DTA systems: analytical approaches and simulation-based approaches. Analytical approaches solve the
DTA problem using mathematical programming. In 1978, Merchant and Nemhauser were the first to
formulate a DTA system as a mathematical program (3). Their formulation represented a deterministic,
fixed-demand, system optimization problem. Their efforts were followed by several attempts to model
DTA as a mathematical problem, such as Janson who in 1991, proposed a User Equilibrium mathematical
solution for a DTA problem.

In 2000, Ziliaskopoulos introduced a linear programming formulation for a system optimal DTA problem
(3). His formulation was based on the cell transmission principle which added more response to traffic
conditions. Using mathematical programming to formulate DTA was limited by constraints related to a

7
trade-off between mathematical traceability and providing efficient solutions for real world large-scale
traffic networks (3).

Several simulation models have developed based on DTA, including DYNASMART-P, CONTRAM,
DYNAMIT, and TRANSIMS. These models use established analytical formulation and mathematical
programming along with a traffic simulator to reproduce traffic flow dynamics. Simulation-based models
are more widely used and accepted because they provide more realistic traffic representation than
analytical approaches.

DTA simulation models have a traffic simulator, and have the ability to account for the variations in the
O-D trips in real-world networks. DTA simulation models use the current available data to incorporate
real-time variations in the O-D demand, while maintaining the procedure computational efficiency (43).
They have the ability to model congestion during peak and off-peak conditions, in addition to non-
recurrent congestion events by modeling the buildup and dispersion of traffic on network links (34). Also,
DTA simulation models can be used to estimate real-time current conditions on the network, and predict
short-term future network conditions using the rolling horizon framework. Real-time data from
surveillance systems are combined with historical data to estimate the network’s real-time current
conditions and predict short-term future conditions (39).

Conventional tools do not explicitly model driver behavior patterns on the network. Therefore, they are
inadequate for evaluating the effects of operational and traffic management strategies, such as
implementing various intelligent transportation systems (ITS) technologies. In evacuation modeling, the
principle of minimizing travel time still applies to evacuation trips, but there may be some restrictions by
the authorities and roadway conditions that will affect the choice of evacuation trip route.

2.1.2 Evacuation modeling parameters

Evacuation studies have been performed in conjunction with emergency preparedness planning for
various types of disasters. Evacuation planning emphasizes the development of scenarios and selection of
the most appropriate response to that emergency scenario. During the process of developing an
evacuation plan, there are several parameters that need to be understood to develop an efficient plan and
facilitate communication among the professionals and decision makers. These parameters include (40):
1) Protective action recommendation (PAR) is a recommendation from the technical staff working on
developing the evacuation plan to the decision makers. It usually consists of three alternatives for the
at-risk population: do nothing, shelter in place, or evacuate the area
2) Evacuation time estimates (ETEs) is the time period required to move the at-risk population out of
the danger area. These are aggregate measures and are different from individual evacuation times,
which are defined as the time period from the actual beginning of the evacuation trip until the
evacuees are cleared out of the evacuation area
3) Emergency warning time consists of two periods: decision period and dissemination period.
Decision period is the time decision makers need before they decide to issue the emergency warning,
while the dissemination period is the time needed for the warning to reach the public
4) Mobilization time is defined as the time period needed by the evacuee to prepare for evacuation
before the beginning of the actual evacuation trip. The time needed to prepare for the evacuation is
related to the level of risk and the available warning time. In emergency events with short warning
times, the main factor in determining the departure time for the public is when they receive the
warning. The time people need to mobilize is the main factor in determining their departure time for
emergency events with long notification periods (23)
5) Clearance time is defined as the time elapsed from when warning was received to when the evacuee
reached the evacuation trip destination. Clearance time is usually short for short warning time events

8
 (less preparation and closer destination) while it is long for hurricane type events (an average of about
nine hours for Hurricane Floyd) (23)

Traffic flow modeling deals with traffic assignment to evacuation routes, congested bottlenecks, and trip
departure timing. In developing traffic evacuation models, an essential element is defining the evacuation
transportation network, link geometry, and capacity. Also, an important issue would be defining the level
of detail required for the network and emergency planning zones (EPZ), which are largely dependent on
the output precision representation and computational speeds desired by the user (28).

Every EPZ should be determined based on event risk and evacuation risk (28). The defined EPZ should
be homogenous in its exposure (vulnerability) to the hazard or event justifying the emergency response
for the entire EPZ. Also, the EPZ should be homogenous in its evacuation risk, making the evacuation
recommendation valid for the entire EPZ. After defining the EPZ, it could be divided into sectors called
emergency response planning zones (ERPZ) based on the population and evacuation route (28). More
people are going to respond to evacuation instructions if they know they are in an evacuation zone (28),
which is why ERPZ should also be defined by political borders (city or county limits) and well
recognized geographic features (rivers) or street networks (the interstate system). If the public is unable to
identify the boundaries of the EPZ, that could lead to either people not evacuating when they need to,
putting them in danger, or too many people evacuating from zones when they did not need to, adding
unnecessary strain on the transportation system and adding more to the ETE.

The EPZ could be based on a single fixed point from which the emergency area radiates, as is the case in
manmade emergency events like chemical spills or nuclear power plant related emergency events. Also, it
might be variable in size, source, and direction, as is the case in natural disasters like storms or wild fires
(7). The shape and size of the evacuation area are determined by the size and growth rate of the source
itself, and they could be classified as centralized or regional. Fixed sources that expose the population to
danger in their immediate surrounding are labeled as centralized, while evacuation areas that are not only
defined by the emergency source are labeled as regional (7). The rate of growth of the evacuation area is
mainly a function of the nature of the emergency itself and the speed by which it moves (7). The size and
characteristics of the evacuating population are determined by the size of the evacuation area, the social
and economic attributes of the population, and the level of danger to the public. The level of threat to the
public safety may vary with time or remain constant, but it is a major factor in determining how quickly
the population needs to be removed from the danger area (7).

2.1.3 Modeling tools for evacuation planning

In the research related to emergency evacuation modeling, there are two main topics of special interest:
human behavior and computer simulation (35). These two main areas of evacuation modeling could be
combined at three different levels: evacuation time estimates (ETEs) analysis, collection of background
data, and human behavior analysis (28). Computer models provide relatively low-cost, low-risk tools for
understanding the influencing factors in emergency evacuations and preparing evacuation plans relative to
the cost of not having a plan prepared when one is needed. They are mainly used to test various
assumptions and analyze the results for different alternatives.

A limitation in evacuation models is that they were focused on developing new traffic flow models while
few of them considered the background data and behavioral aspects analysis. Ignoring the human
behavior aspect when modeling emergency evacuation could lead to underestimating ETEs.

Early traffic evacuation models aimed at obtaining the ETEs for different scenarios were mainly based on
roadway capacities and traffic demand. In most of the literature dealing with evacuation modeling, the
capacity of the evacuation network was determined based on Highway Capacity Manual (HCM)

9
procedures. The limitation to this approach is its inability to account for network capacity changes during
emergency events (28). Also, it is generally not suited for use with over-saturated traffic networks, such
as during emergency evacuations.

Simulation models based on DTA algorithms can handle both traffic dynamics and travel behavior. In
addition, they provide an effective platform for evaluating different operational strategies and analyzing
their performance and benefits. Real-time transportation simulation models developed over the past
decade enabled traffic managers and engineers to better manage and direct traffic, but those models
require more detailed input data and more intensive computer modeling.

Different levels of traffic representation are used by different classes of simulation models. In
microscopic simulation models, the interactions of individual vehicles are captured by using algorithms
that represent vehicle acceleration and deceleration, passing maneuvers, and lane changing behavior.
Conversely, macroscopic simulation models do not capture the movement of individual vehicles on the
transportation system. Macroscopic models are used to simulate the traffic flow based on speed, flow, and
traffic density relationships, and do not model the interactions between individual vehicles. Mesoscopic
simulation models simulate individual vehicles on the transportation system, but capture their interaction
using aggregate relationships. In mesoscopic models, the travel times are determined based on simulation
average speeds, which are calculated using speed-flow relationships (18).

KLD associates Inc. conducted a comparison between using macroscopic and microscopic models for
emergency evacuation in a study for the Nine Mile Point nuclear power station (40). The microscopic
model used was WATSim while the macroscopic model used was PCDYNEV. The study indicated that
for large roadway networks, macroscopic models provide good practical accuracy and efficiency. In that
study, the macroscopic model evacuated 4% more traffic than the microscopic model with 5% less
simulated evacuation time. In addition, the microscopic model took 300 times the run-time (7 seconds
compared to over 2,100 seconds) of the macroscopic model (40).

Evacuation simulation and modeling tools play an important role in the pre-planning stage, real-time
evacuation traffic operations, as well as post-emergency analysis stage. Traffic simulation software has
seen an increased use for modeling traffic flow under emergency conditions. Simulation tools assist
emergency managers and transportation officials in the decision-making process by providing important
information about evacuation traffic conditions.

Several models have been developed and/or used to analyze the transportation system in cases of regional
emergencies and to improve the evacuation process; the following is a brief description of some of them:

2.1.3.1 Network Emergency Evacuation (NETVAC): It was developed by Shaffi, Mahmasani, and
Powell at MIT in 1982 mainly as a reaction to the Three-Mile Island incident in 1979. It could be used for
route selection, lane management, and intersection control analysis (9). The major drawback for using this
model in evacuation studies is that it does not model drivers’ behavior and it has no travel demand
component.

2.1.3.2 Mass Evacuation (MASSVAC): It was developed in 1985, and it has been used to test the
operational strategies for hurricane evacuations in Virginia (9). A support system called Transportation
Evacuation Decision Support System (TEDSS) was developed by Hobeika et al. for the development,
analysis, and evaluation of evacuation plans around nuclear power stations.

The system utilizes a database module to store disaster related information, evacuation rules, and the
characteristics of the region and the transportation network. The MASSVAC simulation module includes
several traffic assignment algorithms to obtain the initial decision for evacuation. A graphic display

10
module is used to pass the input and output from the simulation module to the user. Finally, a system
control module is used to manage the previous modules (2).

In this simulation module, vehicles are loaded onto the network through an assignment-simulation
process. The input includes different network geometry parameters, origin-destination (O-D) points, as
well as trip productions at each origin. MASSVAC is a macroscopic model based on User Equilibrium
algorithm and is designed to operate in realtime. It loads the evacuation demand on the transportation
network, determines the best evacuation route, and measures the evacuation time (2).

2.1.3.3 Oak Ridge Evacuation Modeling System (OREMS): It was developed in the mid 1990s by the
Oak Ridge National Laboratory (ORNL). It represents a computer-based traffic simulation model that has
the ability to estimate the time needed for evacuation under different scenarios.

OREMS has a probabilistic model that includes network characteristics. It consists of a pre-processor for
data input, a simulation model, and a post-processor that is linked to GIS for the results output and
simulation. It has the ability to account for driver behavior and weather conditions, but it can only assign
passenger cars since it does not perform modal split (14). Also, it has the ability to be used for real-time
operations, but that requires a large amount of data to be fed into the model and analyzed before the
model becomes operational, which is a time consuming and a labor intensive process (9).

The input data are updated automatically, with the ability to perform DTA in real-time. Trip attractions
are calculated based on experience due to lack of shelter data (14). OREMS was developed for the
Federal Emergency Management Agency (FEMA) to be used in the Chemical Stockpile Emergency
Preparedness Program (CSEPP). It can be used for ETEs, identifying traffic bottlenecks, and for the
evaluation of traffic management strategies (35).

2.1.3.4 Post, Buckley, Schuh and Jernigan (PBS&J) Model: It was developed in 1999 by PBS&J Inc.
as a Web-based evacuation travel demand model (it is also known as an Evacuation Traffic Information
System [ETIS]). Its main function is to monitor major congested traffic areas during evacuations for
hurricane type events (14). The input for the model consists of evacuation data on the county level while
the output represents traffic volumes on the regional transportation network.

The PBS&J model is housed at FEMA’s Region IV (Alabama, Florida, Georgia, Mississippi, South
Carolina, and North Carolina) regional operation center (14). FEMA considers this model a good tool for
evacuation planning. For input data, the model requires roadway network, route, evacuation demand, and
external stations data on the county level, leading to less data and processing time requirements. The
PBS&J model has the ability to account for human behavior and weather data as input. It can also use
real-time traffic data as input to support traffic operations during evacuation (14).

In this model, the trip distribution is performed manually (while most of the models use the gravity
model) using percentages based on historical data at the county level, and does not have the ability to
perform a modal split step (14). In general, this model requires less input data compared with other
models, but its dependence on historical data and manual updates may limit its effectiveness. It has been
applied to the pre-planning analysis stage for FEMA’s Region IV (14).

2.1.3.5 Dynamic Network Evacuation (DYNEV): It is mainly used for the development of evacuation
plans for areas surrounding nuclear power plants, and has the potential to be used for hurricane
evacuation planning. It has been a component of FEMA’s integrated emergency management information
system used by state and federal agencies (14).

11
Input data consists of network, route, evacuation demand, external station data, and it has the ability to
use human behavior and weather data as input. The input data structure is similar to OREMS, with the
addition to bus data, but it cannot handle real-time data. The DYNEV model can provide ETEs, but it
cannot provide real-time output (14).

2.1.3.6 Personal Computer based Dynamic Network Evacuation (PCDYNEV): It is a macroscopic

model which consists of two main parts: integrated TRaffic Assignment and Distribution model (TRAD)
and Interactive Dynamic Evacuation (IDYNEV). TRAD is used for the distribution of trips between the
zones and their assignment to the different links. The input for TRAD consists of the roadway network,
traffic demand and origin points, and the possible destinations for each origin point. TRAD utilizes user
equilibrium assignment algorithm (40). IDYNEV is a macroscopic simulation model that has the ability
to provide traffic flow statistics at different simulation steps for the links in the roadway network (40).

2.1.3.7 Traffic Estimation and Prediction System (TrEPS): Developed in 1995 as a result of DTA
research by the Federal Highway Administration (FHWA), it has the ability to support decision making
for non-recurrent congestion events (e.g., accidents and work-zone) through dynamic traffic estimation.
This model is capable of supporting transportation planning and operation decisions through two major
components: a planning component and a real-time operation component (14). Input data consist of
network, route, evacuation demand, and external station data. It can use real-time traffic data as input to
support traffic operation during evacuation.

2.1.3.8 Network Simulation Model (NETSIM): Developed by KLD Associates Inc., it represents a
stochastic microscopic traffic simulation model (5). This simulation model has the ability to measure the
transportation system performance under different traffic control strategies. It can model both personal
vehicles and buses on congested traffic networks where the demand is processed and analyzed at each
simulation step. NETSIM was applied to estimate evacuation times needed for areas surrounding nuclear
plants. The major limitation to NETSIM is that it cannot handle large regional networks, and it does not
perform dynamic traffic assignment. In addition, it does not have a travel demand component, which
needs to be provided from other sources as input.

2.1.4 Human factors and evacuation modeling

To perform projections of travel demand changes, it is necessary to identify the parameters that influence
the travel decisions of each individual. With the increasing need for travel demand models to perform
more complex and advanced tasks conventional static planning tools fall short of fulfilling this need.
Hence, there is increasing emphasis on enhancing those models by improving their input data in addition
to the development of new modeling procedures. These improvements aim to enhance the ability and
reliability of the models in providing future traffic forecasts to support the decision-making process in
addition to supporting traffic operation analysis.

Actual human behavior patterns, along with the response rates, have major impacts on the effectiveness of
the results from evacuation traffic simulation models. A distinctive feature of human behavior in
emergency situations is that families tend to evacuate as a unit (35), where the driving members of the
family pick up the non-driving members and then evacuate out of the danger area. In evacuation traffic
management plans, authorities should discourage certain movements but not prohibit them. The
application of the evacuation plan should be flexible because the driver might have a strong reason (to
pick up someone or to join family members) to go to a certain destination during an emergency situation
(40).

Murray and Mahmassani (35) integrated the household trip-chaining behavior with the evacuation
simulation model. In their study, they used linear programming to describe the decision-making process

12
for each household in the model (35). They used trip chaining to model household (HH) behavior, an
approach feasible only for small networks with a limited number of zones and HH. However, when
dealing with regional networks, user equilibrium will most likely be used for the assignment.

Human behavior is usually used to assist in determining the number of trips generated and the trip
departure timing. Many of the reviewed models assumed that everyone who was warned about the
emergency is going to evacuate, all registered vehicles will evacuate, or that one vehicle per HH is going
to evacuate (28). Due to the different nature of the emergency evacuation trips, most researchers rely on
data obtained from travel survey to model emergency evacuations.

2.1.5 Travel surveys and evacuation modeling

Travel demand models utilize data provided by travel surveys to establish trip generation, trip
distribution, and modal split relations. Hence, urban areas need travel survey data to estimate and validate
models that are used for their planning activities by using those data to identify existing conditions and
problem areas in their transportation system. It is very important to develop accurate input data to
improve the accuracy and reliability of the model output. However, conducting surveys and updating
survey data is usually faced by budget and time constraints, limiting the abilities of different
transportation agencies to keep the needed data up to date and acquire the data needed for their expanding
modeling needs.

Different methods are used to perform travel surveys. Some of the most commonly used method include
household travel and activity surveys, vehicle intercept and external station surveys, workplace and
establishment surveys, and parking surveys (53). For evacuation modeling applications, travel survey
methods are used to collect travel behavioral aspects in addition to common data about the number of
daily trips, their destinations, and choice of travel mode. Travel behavioral data collected are determined
based on the modeling need, and could include activity based travel summaries, usage of different
vehicles for different trip purposes, and stated response travel behavior.

The process of implementing travel surveys is performed over five major stages: planning, design, field
implementation, data preparation, and data analysis (8). The survey planning stage is mainly dependent
on the scope of the study and the needed data analysis type. In this stage, the problem that needs to be
studied is identified. Also, the hypotheses associated with the relations that need to be examined are
identified.

At the survey design stage, the best method to acquire the data needed is determined where the survey
method, the associated time frame, and budget requirements are established. The different tasks
performed at this stage include collecting background information, survey design, organization, sampling,
drafting, and constructing (8). The design of the survey is based on the data analysis needed for the
project to provide the most efficient method to achieve the objectives of the study.

After the survey design is finished, the survey is then implemented and data collection from the
population sample begins. The results from the survey are coded into the system and the data are tested
and cleaned to keep the usable data. Afterward, the data are processed and prepared in an analysis-ready
format (8). At the data analysis stage, the relations between the response variable and the other variables
are established where the data is tested using different statistical measures. Also, the results and
conclusions of the study are presented and applied to the problem under investigation.

Surveys that are designed to collect data describing actual travel behavior are classified as revealed
preference (RP) surveys (8). On the other hand, stated response (SR) surveys collect data on how the
transportation system users would respond to a hypothetical situation (8). Evacuation modeling usually

13
falls within the emergency planning stage where the incident is hypothetical; therefore, most surveys used
for evacuation modeling are SR surveys.

Concerns exist about the reliability of SR data for evacuation modeling because what people say they will
do under a specific set of circumstances may be different from what they would do if actually faced with
these circumstances (8). A number of studies have been reviewed where the validity of data obtained
using SR techniques was checked, and it was concluded that in most cases the SR techniques can provide
predictions of choice behavior to a satisfactory degree with the need for additional systematic validity
research (8).

There are four general classes for SR survey techniques based on the nature of the questions and the
expected behavioral outcome: stated preference, stated tolerance, stated adaptation, and stated prospect
(8). Stated preference (SP) surveys represent the most important source for developing evacuation models
to capture travel decisions where most of the expected behavioral outcomes and constraints are provided.
In stated tolerance (ST) surveys, respondents are asked to provide the travel conditions that would prompt
them to take a certain action. Stated adaptation (SA) surveys ask respondents to provide relatively open-
ended responses to a set of certain constraints, while stated prospect (SPro) surveys provide respondents
with some general scenario to elicit their constraints and expected behavioral outcomes.

Stated preference methods refer to the techniques of collecting and modeling with data collected in the
form of preferences as reflected in rating, ranking, or choices among hypothetical alternatives
characterized by a set of pre-specified attributes that can take different values (4). SP tools were
developed in the early 1970s for marketing research, but it wasn’t until the early 1980s before the initial
application of SP surveys for transportation planning (8). The basis of these methods is the observation of
behavioral choices by confronting respondents with hypothetical situations. SP techniques are most useful
in transportation planning to measure perceptions and attitudes, and estimation of policy responses,
potential demand, and elasticity for transport-related choice sets, such as different travel modes, vehicle
types, or route choice (4). The travel behavior of each individual reveals a utility function which provides
a means of forecasting future choices. The use of SP techniques has become increasingly common
practice for evacuation modeling studies.

Despite the wide use of SP techniques to support transportation studies, the SP survey designs are still
subject to a range of experimental error not found within RP survey designs (4). The debate about the
predictive validity of stated preference remains the main obstacle faced when using SP surveys for
transportation studies (15). The predictive validity refers to the degree of accuracy a hypothetical
response predicts future behavior. It is important to have correlation between the stated behavior and
actual behavior for the survey to be useful in the analysis (15). The validity of SP data can be maximized
when attention is paid to the hypothetical choice circumstances so they are realistic and relevant to
individual respondents (4), meaning that the SP survey design is very often the most important
determinant of the internal and external validity of any SP data collected. The objective of using
experimental design in SP surveys is to present attributes that are varied independently to the population
sample surveyed so that the effect of the different attributes on the respondent’s behavior is identified (8).
When comparing stated preference data and revealed preference data for the same scenario, they are
unlikely to have the same variation. However, the decision-making process is similar and thus the utility
functions should be proportional to each other with stated preference data being limited by the hypothesis
made (25). Also, well designed SP surveys are capable of producing comparable results with the
independent evidence of the RP models because respondents to a hypothetical choice situation behave in
basically the same manner as those facing real conditions (4).

14
2.2 Evacuation Modeling
Good evacuation planning involves an iterative process to estimate the best evacuation routes and their
corresponding ETEs for different regional emergency scenarios based on local data. The objectives of this
study are achieved in two stages: development of local demand loading rates and drivers’ behavior data
for different emergency evacuation scenarios and incorporating the data into the evacuation modeling
tool.

2.2.1 Evacuation travel demand

A major step in evacuation modeling is identifying the demand for travel on the transportation system.
The evacuation model should help the planner estimate the size of the population that needs to be
evacuated, in addition to the origins and destinations of their trips (7). By taking into consideration the
response rates to evacuation orders and the ability to evacuate, three main categories of the evacuating
population are defined: residents who live in the danger area, tourists and visitors who mainly stay at
hotels and motels, and special facility population, such as those in jails, schools, hospitals, and nursing
homes.

The objective of trip generation is to identify the demand or the number vehicles that are going to be
using the transportation network. The demand for travel is mainly based on the number of households
(HH) and family size since the HH is the basic unit of evacuation. Many factors influence the public
compliance rates with evacuation instructions. Some of them are related to different behavioral aspects,
while others are associated with the nature of the emergency itself.

For evacuation participation rates, most behavioral models rely on historical data, especially data obtained
from hurricane evacuations (14). Sometimes the population evacuates without receiving the instructions
to evacuate from the officials. On the other hand, some segments of the population do not comply with
the instructions to evacuate. The public response rates generally increase when the warnings are
frequently repeated with conformity and when they are issued from a source the public perceives as
credible (23).

Families that have a prepared emergency plan are more likely to comply with evacuation instructions than
those without a prepared plan at the time of the emergency. According to Mileti and Sorensen, the
sequence of public response to emergency warning include receiving and understanding the warning,
believing the warning is credible, personalizing the warning, confirming the warning, and responding by
taking protective action (23). Also, Perry and Lindell defined four stages that define the public response
to emergency warning, which include risk identification, risk assessment, risk reduction, and protective
response (23).

A number of issues have been related to public compliance with instructions to evacuate, such as shadow
evacuations, spontaneous evacuations, and the “cry wolf” effect (23). Shadow evacuation is defined as the
population evacuating from the emergency shadow regions (regions surrounding the evacuation area).
Data from Hurricane Floyd’s evacuation indicated high evacuation rates from low risk shadow areas.
Spontaneous (early) evacuation occurs when the risk seems imminent even before official evacuation
instructions are issued. This is especially noticed in emergencies involving short warning times such as
hazmat spills incidents. The “cry wolf” effect is the noncompliance of the public to the evacuation
instructions because of responding to false alarm messages in the past, which greatly affects the
credibility of the warning messages issued in the future (23).

Different assumptions were reported in the literature regarding evacuation compliance rates. Hobeika et
al. (1994) assumed a 100% response rate without any consideration to evacuation from the shadow region

15
surrounding the at-risk area. Lindell et al. (2002) used regression analysis to develop a function for the
evacuation response rates for hurricane type events. Based on the category of the at-risk area and the
hurricane category (28), he developed the following equation:

𝑌 = 33.78 − 19.72𝑋1 + 2.87(𝑋1 )2 + 31.39𝑋2 − 2.84(𝑋2 )2 Equation 2.1

Where, Y: % compliance with evacuation instructions, and

X1, X2: Risk area category and hurricane category.

In the literature, the size and distribution of the transient population within the EPZ was usually estimated
using the hotel and motel room average occupancy. A compliance rate of 100% with evacuation orders is
usually used for visitors, and there is very little literature available for this population category. An
important factor to be considered is the seasonal variations in average occupancy, which are usually
reported by month. The demand for this category of the population is usually defined as 1.0
vehicles/room, since they are less likely to have more than one vehicle (1). Also, those who travel on
buses are most likely to depart as a group using the same bus, reducing the number of vehicles in the
evacuating traffic stream.

The evacuation participation rates also vary greatly for different types of hazards. Compliance rates are
usually high (could reach the 90% range) in incidents involving hazmat or in major storm surge areas,
whereas for small storms or river floods they are usually low (23). From the literature (primarily from
hurricane type events), an average of 1.3 vehicles per HH were used in the evacuation. People who did
not have access to private vehicles were usually picked up by family or friends, and only an average of 1-
2% percent needed official assistance (23).

Some empirical studies and surveys are available in the literature describing the number of evacuating
vehicles per HH from the EPZ. Cora and Johnston (2002) used the Poisson distribution to model the
number of evacuating vehicles. The results showed a skewed distribution with a mean (λ) equal to 1 and
mode equal to 2 with a range of 2 to 6 vehicles evacuating per HH (28). Ruch and Schumann (1997)
estimated the number of evacuating vehicles to be 1.35 based on a behavioral survey for the study area.
Also, Prater et al. (2000) reported a rate of 1.34 vehicles per HH for Hurricane Bret (28). As for transit-
dependent populations, the size and distribution of that category was ignored in the previous research
since it is tough to account for and it should have minimal effect on traffic congestion during evacuation.

2.2.2 Evacuation trip loading rates

Evacuation trip loading rates have a major effect on the transportation system congestion levels,
operational conditions, and clearance time. Residential HH departure time distributions determine the
demand rate loading of evacuating vehicles onto the available evacuation transportation network.
However, data obtained from the nationwide census do not cover all the information needed to develop
estimates of the mobilization time, and may not specifically focus on the area to be evacuated. In such
cases, telephone surveys have been used to obtain information about families and estimates of the
response time to certain threat scenarios (40). Analysts agree that the correct function to describe
departure time in emergency evacuations is a sigmoid curve (28). The sigmoid
function is a special case of the logistic function (52, 51), which is as follows:
1
𝑃(𝑡) = Equation 2.2
1+𝑒 −𝑡

Where: P(t): Population percentage departing at time (t), and

t: Time elapsed since the beginning of the evacuation process.

16
The logistic function is the inverse of the natural logit function and it is used to convert the logarithm of
odds into probability (52). The curve shows exponential growth for negative “t” that slows down to linear
growth near “t = 0” and approaches “y = 1” with an exponentially decaying gap.

Cova and Johnson (2002) used a Poisson distribution to model the overall departure times for evacuating
vehicles using five-minute increments (28). Radwan et al. (1985) used a sigmoid curve with the following
formula:
1
𝑃(𝑡) = Equation 2.3
[1+𝑒 (−𝛼(𝑡−𝛽))

Where: β: median departure time, and

α: slope of the curve.

Also, Tweedie et al. (1986) developed his function based on data obtained from local civil defense
officials. He generated a distribution function (28) using the formulation:

𝑡
�−0.5� 2 ��
𝑃(𝑡) = 1 − 𝑒 𝛽 Equation 2.4

Where: β: Mode of the distribution function.

Hobeika and Kim (1998) used an exponential function to describe departure times (2). Their function was
equivalent to the function proposed by Radwan et al. (1985) with β = 45 minutes. Lindell et al. (2002)
combined warning and preparation time distribution and analyzed data collected from coastal region
residents about the time needed to leave their homes after hurricane evacuation instructions were issued
(28). Figure 2.1 illustrates the trip loading curves developed by different researchers for different case
studies.

Figure 2.1 Evacuation departure times curves (28)

Another factor that needs to be considered is the time when the evacuation orders are issued. According to
Klepeis et al. (2001), over 90% of the at-risk population is usually at home between 10:00 P.M. and 6:00
A.M., while 30% of the population is indoors between 10:00 A.M. and 3:00 P.M. This indicates that
emergency planners need to incorporate additional time during the day for adults to arrive home from

17
work and kids arrive home from school, especially in cases where there is a very short warning time, such
as hazmat spills.

2.2.3 Selection of evacuation modes and routes

Data available from literature indicate that the nature of emergency evacuations do not allow for balanced
traffic demand where travelers learn from their mistakes. Several factors impact travelers’ choice of
evacuation route. Evacuees could pick the closest exit point outside the at-risk area, the route with the
least travel time, or the route more familiar to them. Most of the population pick the house of a relative or
a friend, or a hotel room as an evacuation destination (22, 23). In the literature, it is reported that on
average, 13% of the population will evacuate to public shelters based on the severity of the event and
income levels (23, 49). Hobeika and Kim (1998) utilized User Equilibrium assignment algorithm in
MASSVAC 4.0, where evacuees were assigned to routes that minimize their travel time (2). Cova and
Johnson (2002) assigned traffic to routes based on the shortest travel distance to exit the risk area (28).

Several transportation modes are available for evacuating the public, including private vehicles, buses,
trains, and air. Most of the available literature focused on private vehicles as the primary mode evacuation
since it is the one mode that would significantly impact the surface transportation network. The problem
with flying is that airlines may discontinue flights going into the danger zone even if the local airport is
still open. Also, if the ground personnel evacuated early, that could stop evacuees’ movement through the
airport (28). As for evacuation by buses, they are mainly used for the transit dependent (no access to
private vehicles) and special facility population. Some members of these populations evacuate with
friends or family and it is unclear how that would affect the evacuation of private vehicles. In addition,
using buses for evacuation is greatly affected by the availability of buses, time needed to load and
mobilize those buses, and evacuate outside the danger area. Using transit might extend the ETE time
needed by personal vehicles for two reasons: the need for routes for their return trip may limit the ability
to apply contra-flow strategy, and their presence in the traffic stream and their stops may interrupt the
flow of the traffic stream.

2.3 Description of Software Tools

This study will mainly use two software packages; Cube-Voyager and DYNASMART-P. These two
software packages are used to develop the main features of the model, in addition to other software
packages that are used for data compilation, analysis, and other processing tasks. The following sections
describe the main features and components of the software.

2.3.1 DYNASMART-P software

In 1992, Peeta and Mahmassani developed a DTA model called DYNASMART (DYnamic Network
Assignment-Simulation Model for Advanced Road Telematics) (3). This model has a traffic simulator
with fixed departure times for the O-D demand (it assumes a previous knowledge of the O-D demand for
the planning period). The rolling horizon approach was used to account for the variations in the O-D trips
in real-world networks (43). This approach hinges on the principle of using the current available data to
incorporate real-time variations in the O-D demand while maintaining the procedure computational
efficiency. It provides an efficient way of handling problems that require knowledge of future demand for
the full planning horizon (43). It uses the available information about current and short-term future (5 to
15 minutes) network conditions to determine real-time suitable operations control strategies while
preserving computational efficiency.

DYNASMART-P is an analysis tool that incorporates different information supply strategies, route
assignment rules, and traffic control measures to best fit the functional requirements of Intelligent

18
Transportation Systems (ITS) applications. It combines the principles of traffic assignment and simulation
to overcome the complexity usually involved with DTA problems, and produces a system with adequate
practicality. There are three major components for DYNASMART-P: the simulation component, the user
behavior and information supply strategies component, and the path processing component (20).

2.3.1.1 Simulation component: DYNASMART-P utilizes mesoscopic simulation models to represent

traffic interactions (38). It efficiently tracks the movement and location of individual vehicles throughout
the planning horizon, but it does not monitor microscopic details such as car following (20).

There are two main modules involved with traffic simulation in DYNASMART-P: link movement and
node transfer modules. The link movement module processes vehicle movements on network links by
evaluating links speeds based on speed-density relationships for each simulation step. The node transfer
module determines the link to link (or section to section) traffic transfer based on the control type at the
intersection. This module determines the number of vehicles entering and exiting the network, and
vehicles still in queue at each simulation step.

2.3.1.2 User behavior and information supply component: Traveler behavior modeling in
DYNASMART-P is based on boundedly-rational behavior. That is, drivers are going to alter their trip
route only if they experience a gain over a certain threshold perceived to be sufficient for them.
Information supply systems will provide drivers with indications of travel times over alternative routes,
and sometimes the best route for their trip. Nevertheless, drivers’ degree of response to the provided
information will vary according to their different user class, indifference band, and resistance to switching
routes (20).

2.3.1.3 Path processing: This component determines the impedance of using a specific route using the
traffic attributes obtained from the simulation component following the procedures illustrated in Figure
2.2. For multiple user classes, the algorithm for calculating the K-shortest path is combined with the
simulation model to calculate the (K) different paths for the O-D pair. To achieve computational
efficiency, the (K) shortest paths are calculated at pre-specified intervals instead of every simulation time
step. The shortest path calculations in DYNASMART-P are based on generalized link impedance from
both travel time and out-of-pocket cost. The user cannot assign more than one cost for different links of
the same type in the current version of DYNASMART-P (20).

DYNASMART-P also requires the user to specify the number of shortest paths to be calculated, which is
determined by the planning application need to consider alternate paths for it. Usually a single path
calculation is used for UE or SE applications, two-path calculations for applications with advanced
traveler information systems (ATIS) strategies, and three-path calculations for en-route information
planning applications.

19
Figure 2.2 DYNASMART-P assignment procedure (20)

2.3.1.4 Model features: DYNASMART-P is capable of handling urban traffic networks with various
sizes, up to 89999 nodes (20). In order to achieve a more realistic representation of the network and traffic
conditions, DYNASMART-P allows using restrictions on vehicle movements that impact route
assignment. These restrictions are used in flow simulation, path processing, and provision of information
in the model.

DYNASMART-P can model link intersections with no-control, yield signs, stop signs, and signalized
intersections. It can also model different freeway ramp metering strategies. To calculate the number of
vehicles that traverse the intersection at each simulation period, DYNASMART-P applies outflow and
inflow capacity constraints. Outflow capacity constraints determine the maximum number of vehicles that
are allowed to leave each approaching lane at an intersection. Inflow capacity constraints determine the
maximum number of vehicles allowed to go into the link through a simulation step. Left turn capacity is
based on default values provided in HCM 2000.

Most of the common microscopic simulation models use the critical gap acceptance theory and car
following technique to estimate the capacity of an intersection with yield sign control (20). In
DYNASMART-P, however, for yield, two-way stop controlled, and all-way stop controlled intersection,
vehicles are discharged according to the traffic flow rate on the major approach and the type of turning
movement. The user can either specify the discharge rates or use the values from NCHRP (Capacity and
Level of Service at Unsignalized Intersections, April 1996), which were used for the purposes of this
research.

DYNASMART-P can model pre-timed and actuated signal control, but not at a microscopic level. For
pre-timed signals, DYNASMART-P requires the offset, number of phases, and permissive movements for

20
each phase, and the yellow and green times. The user needs to specify the phases and movements
associated with each phase, along with the maximum and minimum green times for actuated signal
control. DYNASMRT-P is capable of modeling five types of driver classes with different assignment
rules based on drivers’ knowledge of the network, the availability of ATIS, and how users respond to the
provided routing information (20).

2.3.1.5 Demand representation: One of the unique aspects of DYNASMART-P is that it allows the user
to specify generation links for each TAZ. These links are used to load the demand and generate vehicles,
instead of the conventional method of generating traffic on centroids as in most available planning
software. This feature overcomes the problem of generating unrealistic traffic flow patterns around origin
and destination nodes (20).

Generation links can be physically contained only in one zone, even if they are on the boundary of more
than one zone. However, a generation link may be specified to receive demand from more than one zone.
In that case, traffic generated on that link is either proportional to its lane-miles or specified by the user as
a loading weight which determines the share of demand a TAZ will provide. A virtual centroid is
internally generated for each TAZ or for each aggregated zone. Centroids are connected to the destination
nodes where vehicles exit the network. A destination node can be specified for a maximum of two zones
(20).

DYNASMART-P provides some flexibility in the way it accepts vehicle generation information (O-D
matrices). It offers two options for loading demand information. For entering time interval O-D matrices,
the number of loading intervals must be specified along with the start and end of vehicle generation times
for each interval. A multiplication factor for demand generation is specified to facilitate the application of
different levels of demand loading. The total number of vehicles in the network is obtained by adding up
all the entries from all of the specified O-D matrices, and multiplying that sum by the multiplication
factor. Truck and high occupancy vehicles (HOV) demand is specified as an independent O-D matrix in
the truck demand data file or as a fraction of the demand to be loaded onto the network.

For activity-based O-D matrices, the user specifies the vehicle characteristics and travel plan with
intermediate stops and activity duration to model trip chains. It is important to list the vehicles in the
vehicle-trip input file in the order of their departure or in the order of their generation links for vehicles
with the same departure times. There are no restrictions on the number of intermediate stops that a vehicle
makes at intermediate destinations, but DYNASMART-P only reports up to three destinations in the
summary statistics file. This type of vehicle loading is needed for evaluating different traffic management
strategies that require a specific loading pattern and/or fixed vehicle paths over the planning horizon. If
the user specifies the multi-user class (MUC) as a percentage of the vehicle fleet, then this type of loading
is not allowed (20).

2.3.1.6 Traffic-flow model: In DYNASMART-P, the traffic flow and traffic relations on the network are
described using two types of modified Greenshield’s models for traffic propagation (20). For freeways,
where the capacity is larger than arterials, DYNASMART-P uses a dual-regime model. A constant free
flow is specified for the uncongested free-flow conditions, and a modified version of the model is
specified for congested conditions. Figure 2.3 illustrates the functions used in DYNASMRT-P for
freeways. The reason for using this model is that freeways can accommodate dense traffic at near free-
flow speeds.

21
Figure 2.3 Type 1 modified Greenshields’ model (20)

A single-regime model is used for arterials, where the presence of traffic control and intersections
limit the possibility of free-flow speeds (Figure 2.4). In addition, due to the arterials’ limited
capacity, any addition to the traffic volume is going to affect the speed over them. Parameters for the
dual-regime Greenshields’ model in DYNASMART-P were calibrated for the San Antonio (Texas)
freeway system. Single-regime parameters were calibrated for the Irvine (California) surface street
network (20).

22
Figure 2.4 Type 2 modified Greenshields’ model (20)

Equations 2.5a and 2.5b represent type 1 Greenshields model, and Equation 2.6 represents type 2
Greenshields’ model used in DYNASMART-P.

vt = uf 0<ki<kbreakpoints Equation 2.5.a

vi – v0 = (vf-v0).{1-(ki/kjam) ^α } kbreakpoint<ki<kjam Equation 2.5.b
vi – v0 = (vf-v0).{1-(ki/kjam) ^α } Equation 2.6

Where, vi = speed on link i,

vf = speed intercept,
uf = free-flow speed on link i,
v0 = minimum speed on link i,
ki = density on link i,
kjam = jam density on link i,
α = power term, and
kbreakpoint = breakpoint density.

2.3.2 Cube voyager software

The O-D input matrix for the DYNASMART-P is produced using Cube software, a product from Citilabs.
It was used to build the Fargo-Moorhead travel demand model, which is administered by the Fargo-
Moorhead Council of Governments (F-M COG), and it is run using the TP+ component.

23
 2.3.2.1 Model features: Cube software represents a transportation analysis and forecasting system that
integrates modeling and graphical methods to study the transportation system (10). The graphical
representation in Cube is based on Viper (Visual Planning Environment), and together (Cube and Viper)
provide the different editors needed to perform the different transportation planning functions (10).

The software also supports some geographic information system (GIS) features, giving the user the ability
to build the model network from GIS shape files to better integrate the modeled transportation network
with the GIS networks. Cube Base represents the user interface for the Cube system. It provides
interactive data input and analysis, GIS, model building, and documentation.

2.3.2.2 Travel forecasting model: Cube software incorporates several algorithms that implement a
traditional four-step travel demand modeling process. TP+ is among the most common algorithms used,
and is used for the Fargo-Moorhead travel demand model. TP+ may generally be viewed as a processor
which implements user-defined travel demand relationships using what Cube refers to as scripts. These
scripts, for example, are used to implement trip generation rates in the model using local data and
parameters.

To graphically view the transportation planning modeling process, Cube software utilizes a tool called the
Application Manager. It provides a flow chart describing the flow of data from one process to another in
the system. It also clarifies the different parts of the data that are going to be used as either input or output
in those processes. The Application Manager also provides the user with the option of running the whole
application or running only specified steps within the application to save time and serve various analyses.

Traffic is assigned in Cube based on a generalized cost function (which is set to be equal to the travel time
if no other parameters are used) using the modeled network and the trip matrix as input. The software also
offers the option of performing a select-link analysis, which identifies how each O-D pair contributes to
the traffic volume using a specific link.

24
3. METHODOLOGY AND DATA ANALYSIS
This section provides a description of the methodology used to develop the hybrid evacuation model. It
provides a discussion of the methodology used to model an evacuation scenario and the different steps
needed to achieve the objectives of this study. It provides a summary of the data collection, survey
population, and sample selection, in addition to the design of the survey used to collect the data needed
for building the model. Also, this section provides a detailed description of the data analysis used for
evacuation trip generation, trip distribution, and trip loading data. The modeling approach and data
analysis discussed in this section are related to the case study used for this study, which is described in
detail in Section 4.

3.1 Methodology
The performance of the transportation system in cases of regional emergencies has a significant effect on
public safety. Most of the applications of security planning are on the operational level of the
transportation infrastructure with modest consideration of the security aspect in the development of
transportation improvement plans (TIPs). To introduce the transportation aspect into the developed
evacuation plans, traffic simulation tools are used to model the transportation system performance during
emergency evacuation events.

Simulation-based models are among the most powerful tools that could be applied for capturing driver-
network interactions. The objectives of this study included developing an evacuation model that
recognizes MPO resources and data availability. The developed model should have the ability to address
changes in travel demand, perform dynamic traffic assignment, and be capable of evaluating the
effectiveness of different traffic operation measures. In addition, this study aimed at generating
evacuation related data for small- and medium-size MPOs that could be applied to similar areas.

A major component of this study focuses on the development of a dynamic mesoscopic hybrid
simulation. In addition, a public survey was utilized to collect the data needed for developing the
evacuation model. The following sections provide a detailed description of the procedure followed in
developing the modeling tool and the needed data.

3.1.1 Modeling approach

To satisfy the objectives of this study, the developed modeling approach should have the ability to
estimate the evacuation demand, and the time needed for evacuation, and provide system performance
measures for different scenarios. The modeling approach utilized for the purposes of this study was
developed in two stages: the development of the transportation system and the development of the
evacuation modeling tools. The transportation system includes both the demand and supply sides. The
characteristics of demand for travel over the transportation system are determined based on the trip
generation and distribution data used in the model, in addition to the trip loading rates. The supply is
represented by the modeled roadway network and the associated intersection traffic control, lanes, and
speeds needed to determine the capacities of the transportation system to accommodate the demand for
travel.

Due to the complex nature of evacuation modeling and traffic management during evacuations, both
DYNASMART-P and Citilab’s Cube software were used. DYNASMART-P (Version 1.3.0) simulation
software was selected because it provides connectivity with UTMS, thus making it efficient to obtain the
demand for travel and update the models to reflect new conditions in the field. In addition, it is based on
DTA, making it suitable for modeling evacuation scenarios which are dynamic by nature. Also,

25
DYANSMART-P represents a mesoscopic simulation class model that could be used to model traffic on a
regional level with reasonable input data requirements while providing the needed level of detail for the
purposes of this study.

To develop the hybrid approach utilized for this study, Cube software was used in addition to
DYNASMART-P. The Cube software provides a graphical view of the transportation model structure
through the Application Manager, which represents a flow chart describing the flow of data from one
process to another in the system. In addition, it specifies the different parts of the data that are going to be
used as either input or output for different scenarios, making it efficient to manage the data. To build the
hybrid model, an interface was developed that enabled the exchange of data and output between Cube and
DYNASMART-P software.

3.1.2 Model development

The modeling development primarily consisted of integrating a mesoscopic simulation model

(DYNASMART-P) with the regional travel demand model developed in Cube software. The modeling
approach utilized for the purposes of this study is developed in several stages. They represent the different
steps needed for model preparation and calibration before the model could be used to assist in developing
evacuation plans. The following sections summarize the different stages of the modeling approach
development.

3.1.2.1 Data Collection: Two types of data were needed to develop the evacuation planning model: data
for estimating daily travel demand, and data needed for evacuation modeling. The data needed for
regional travel demand modeling were provided by the Census for Transportation Planning Package
(CTPP), or through local agencies and various data collection efforts. On the other hand, the data needed
for evacuation modeling were not available in the census or the literature and had to be collected through
a public survey. A more detailed description of the data needed for model development and evacuation
modeling is provided in model calibration and data analysis sections.

3.1.2.2 Model Preparation: The study area was divided into TAZs that were homogeneous in their trip
generation characteristics. After that, the roadway network was developed as a set of links and nodes and
their corresponding number of lanes, speeds, direction, control, capacity, and other properties.

After the regional travel demand model was built to represent daily traffic, a corresponding mesoscopic
simulation model was developed. The simulation model incorporated network geometry and social and
economic data. In addition, the signal timing plans were also used as input to account for the traffic
operations side of the simulation. The two models were integrated to form the hybrid model using an
interface developed for the purposes of this study to facilitate the exchange of data between the two
models. This interface prepares the O-D matrix from Cube in the desired format for DYNASMART-P
input file. It also prepares the output link travel time from DYNASMART-P to be fed back into Cube.
The different steps involved in model preparation are discussed in more detail in the hybrid model
development section.

3.1.2.3 Model Calibration: The calibration of the model used for the purposes of this study was
performed in two stages: calibration of the regional travel demand model and calibration of the hybrid
model. The regional travel demand model was calibrated to reflect travel conditions on the transportation
system during an average week day. For the calibration of the hybrid model, the travel time over the links
was checked to test the stability of the model runs. Several runs were conducted until convergence in link
travel times was achieved, and the overall model was checked in terms of the acceptable levels of error,
where the O-D matrix loading rates were adjusted to improve the model performance. Next, the number

26
of simulation runs required was determined based on the variance of the performance measures from
simulation results.

3.1.2.4 Model Runs: After the hybrid model was developed and calibrated, hypothetical emergency
scenarios were incorporated within the model. The hypothetical scenarios selected for the purposes of this
study were determined in consultation with local emergency managers and transportation officials. For
each emergency scenario, the location, affected area’s shape and boundaries and available warning times
were determined. To incorporate the hypothetical emergency scenario within the hybrid model, the
necessary revisions to the transportation network were made, and the input O-D matrix and travel demand
data were modified for each scenario. Also, traffic control parameters were changed to reflect any traffic
management plans used for a particular scenario if applicable. After that, selected system performance
parameters were calculated to compare the different scenarios and traffic management options if
available.

3.1.3 Daily travel demand data

Traditional travel demand modeling utilizes the Urban Transportation Modeling Systems (UTMS)
procedures. The inputs for UTMS involve specifying the characteristics of the activities generating
vehicle traffic on the transportation system, while the outputs represent the estimated vehicle traffic flows
on the system generated by those activities. The data collected were used to develop models and functions
that relate travel behavior to attributes that could be directly forecasted. The data needed for modeling
daily travel demand were available either through the CTPP, or developed by local agencies.

The data needed consisted of the transportation network data, the different social and economic data of
the modeled area, and the data needed for calibrating and validating the developed model. The network
data consisted of the geometry of the roadway network in the area, number of lanes, traffic control,
speeds, functional classification, and other attributes needed to represent the roadway network. The social
and economic data collected represent the data needed to develop trip generation and distribution models,
and were grouped for each traffic analysis zone.

The social and economic data for each TAZ included the number of households, household size, school
enrollment, number of jobs by category, and population by age group, in addition to other data needed for
model development. As for the data needed for model calibration and validation, they included traffic
counts on the major roadways, the number of vehicle miles traveled in the area, trip length distribution,
and counts of the number of daily trips crossing the screen lines in the modeled area. After the needed
data were either obtained or developed, the process of developing a calibrated regional travel demand
model began.

3.1.4 Regional travel demand model

Traditional travel demand modeling is based on the UTMS procedures. The UTMS is utilized to predict
the number of trips made, their time of day, trips origins and destinations, the mode of travel used, and the
routes used for those trips in the metropolitan area (29). In order to model the study area the
transportation network was built, and the study area was divided into TAZs. The TAZs represent the basic
units used for estimating trip productions and attractions in the regional travel demand model. The TAZs
were used to group areas that have similar social and economic attributes related to trip generation in the
study area. For the purposes of this research, the modeled area was divided into 543 internal TAZs. The
boundaries for these TAZs were defined by major roadways in the modeled metropolitan area in addition
to natural barriers prohibiting the movement of traffic. The case study used for the purposes of this
research was the Fargo-Moorhead metropolitan area. Figure 3.1 provides a map of the TAZs representing
the modeled area.

27
Figure 3.1 The modeled area traffic analysis zones

The transportation network was represented as a set of links and nodes that were assigned different
properties, such as number of lanes, speeds, direction, control, capacity, functional class, and turning
lanes. There were several steps involved in the process of constructing and calibrating the regional travel
demand model; those steps included data preparation, trip generation, trip distribution, mode split, traffic
assignment, and model calibration. Figure 3.2 shows the different steps involved in traditional travel
demand modeling.

28
Figure 3.2 The four-step travel demand model

Data preparation was performed before the modeling process started, which included developing the
transportation network from geographic information systems (GIS) format and assigning the different
parameters to the links, nodes, and TAZs. Another component in the data preparation step was the
capacity calculations, where the network links were assigned hourly capacities based on their different
attributes. The Highway Capacity Manual (HCM) procedures are usually used in travel demand modeling
to determine the capacity and measure travel delays. However, this method has several drawbacks when
used for travel demand modeling applications. The HCM procedures depend on the traffic volume and
turning percentages for intersection analysis, which are dynamic, making it difficult for the model to
converge on a solution (18).

The regional travel demand model utilized HCM capacity equations for rural interstate highways that are
based on the number of lanes, percentage of trucks, and speeds (47). For roadways in urban areas, the
procedure in chapter ten of the National Cooperative Highway Research Program (NCHRP) report 365
was used (32). The NCHRP capacity calculation are based on the functional class, number of lanes,
intersection configuration, left turn lanes, and right turn lanes. Table 3.1 summarizes the capacity values
used for the F-M COG travel demand model (1).

29
Table 3.1 Modeled capacities for urban and rural roads
Capacities (Vehicle/Hour/lane)
Each
Functional Multi Lane Each Right Each left
One Lane Additional
Class (Per Lane) Turn Lane Turn Lane
lane

Interstate - 1,800 - - -
Rural

Non-Interstate 1,500 1,700 - - -

Interstate - 1,700 - - -

Major Arterial/
1,000 - 800 300 75
Urban

One-way

Minor Arterial 675 - 600 200 75

Collectors/
450 - 400 100 75
locals

After the data preparation step was performed, the next step was trip generation. Trip generation uses
social and economic data to predict the number of trips produced by and attracted to each zone within the
study area. A trip is defined as the movement between a single origin and a single destination for a single
purpose (29). Trip generation rates were based on historical data to estimate the number of trips generated
to participate in different activities. The Institute of Transportation Engineers (ITE) Trip Generation
Manual provided the needed statistical data and regression models that were used for trip generation (21).

Trip productions in the regional travel demand model were based on the number of single-family and
multi-family dwelling units which were obtained from the 2000 census data, in addition to building
permit data for the years 2000 to 2005 provided by the F-M COG. The model was used to estimate the
number of home-based work (HBW), home-based other (HBO), and non home-based (NHB) trips
produced by each TAZ. In order to develop trip attraction estimates, the TAZs were classified as either
being within the central business district area (CBD) or within a non central business district area
(NCBD).

For university trips generation, major universities in the metropolitan area were treated as special trip
generators in the regional travel demand model (1). To account for school trip generation, the population
was divided into two different age groups to distinguish between high school- and grade school-aged
students for estimating the home-based school (HBS) attraction trips for each category.

In travel demand modeling, the total number of trips produced must be equal to the total number of trip
attractions where each production must be coupled with an attraction to form a trip. Trip productions and
trip attractions were calculated separately; hence, adjustments were made to make the total productions
match the total attractions. Because trip production models provide better estimates of trip rates than trip
attraction models, Equations 3.1 and 3.2 are used for trip generation adjustments (29).

30
CTP = ∑Pz + ∑Pe - ∑Ae Equation 3.1

Where: CTP = Control total of productions,

∑Pz = Trip productions for each zone,

∑Pe = Trip productions at each external station, and

∑Ae = Trip attractions at each external station.

F = CTP / ∑Az Equation 3.2

Where: F = Adjustment factor,

CTP = Control total of productions, and
∑Az = Trip attractions at each zone.

The factor resulting from this process for each trip purpose was applied to each TAZ’s attraction total to
provide the new adjusted attraction values and match the total trip productions and attractions.

After the trip generation step was finished, trip distribution models were used to establish the flow of trips
from production zones (trip origins) to attraction zones (trip destinations). The most commonly used type
of trip distribution model is called the gravity model. In the gravity model, the number of trips between
zones is based on the levels of activity and relative attractiveness of the zones. To account for the
impedance to travel, friction factors were used to make shorter trips more desirable than longer ones.
Friction factors represent an inverse function of the cost of travel between the trip origin and destination
zones.

3.1.5 Regional travel demand model calibration

After the process of developing the regional travel demand model was completed, the calibration and
validation of the model began. The calibration of travel demand models is vital for accurately modeling
current and future travel patterns in a metropolitan area. In model calibration, the different modeling
parameters were adjusted until the predicted travel behavior matched the observed travel behavior for the
base case. The process of calibration and error checking is performed at each modeling step to minimize
the errors in the overall model. Figure 3.3 illustrates the procedure followed for calibrating the regional
travel demand model for the purposes of this research.

31
Figure 3.3 Regional travel demand model calibration

32
The first step in the calibration of the regional travel demand model was to check how the trip length
distribution compared with the length of trips provided by the CTPP. The friction factor coefficients were
used to make shorter trips more desirable. The length of HBW, HBO, and NHB trips were compared with
data obtained from the census, and friction factor coefficients were adjusted until the two data sets
matched.

After the modeled trip length distribution matched the census data, the vehicle miles traveled (VMT) were
calibrated. The VMT in the regional travel demand model are a function of the number of trips generated
in the model and their corresponding trip lengths in miles. To calibrate the VMT values, the total VMT
values for the entire metropolitan area were checked against the reported VMT values. The trip generation
rates were adjusted to change the total number of trips generated on the transportation system until the
modeled and reported VMT values were similar. After the total VMT values were adjusted, the modeled
VMT values by roadway functional class were calibrated to match the reported values. If the modeled
VMT did not match the reported VMT for a certain roadway functional class, then the global speeds and
node delays were adjusted until they were within criteria.

To calibrate for trips distribution between different parts of the modeled area, screenline counts were
used. Screenlines are long imaginary traffic analysis lines that bisect the entire modeled region, and they
are usually represented by major roadways or geographic barriers. For the case study, the screenlines used
included I-29, I-94, the Red River, and the main railroads tracks. If the total modeled traffic screenline
volume was above the specified criteria, a lower k factor was assigned to inhibit traffic from crossing the
screenline. Similarly, if the screenline had a total modeled traffic volume below the designated criteria, a
higher k factor would be applied to the affected zones. This made zonal pairs that cross the screenline
more attractive. After achieving an accurate screenline distribution, the calibration process was repeated,
starting with checking the trip length distribution, until all the successive calibration components were
completed.

The last stage of the regional travel demand model calibration is to check the roadways’ modeled average
daily traffic (ADT) against the actual traffic count data from the field. If the difference between the
modeled and observed traffic volumes was significant, then global speeds were adjusted based on the
area’s land use characteristics. The root mean square error (RMSE) was then used to determine the
overall difference between modeled and observed daily traffic volumes on the roadway links. Also, the
coefficient of correlation (R2) was used to check for correlation between modeled and observed roadway
traffic volumes.

3.1.6 Development of the mesoscopic simulation model

Following the development and calibration of the regional travel demand model, a corresponding
mesoscopic simulation model was prepared for the modeled area. The mesoscopic simulation model
(DYNASMART-P) was integrated with the regional travel demand model developed in Cube software.
The simulation model used network geometry data, and the demand data represented by the O-D matrix
from the regional travel demand model. In addition, the traffic signal timing plans were used as input data
to account for the traffic operations characteristics in the modeled area.

The network geometry data included the roadway number of lanes, turning lanes, free-flow speeds, and
other attributes. These data were developed earlier for the travel demand model and were exported
directly from the GIS data base into DYNASMART-P using the DYNASMART-P editor (DSPED). The
control data consisted of the locations and types of traffic control devices in the modeled regional
network, in addition to the traffic signal timing plans. The locations and types of traffic control devices
were developed from the GIS data base using DSPED, while the signal timing plans were coded manually
after being provided by local transportation agencies. The demand data for the simulation model consisted

33
of O-D tables from the regional travel demand model. The traffic volume data from several counting
stations were studied to determine the loading rates for the simulation model.

After the simulation model was developed, the model was checked for errors that may occur in the coding
of the demand, network data, and model parameters. The process of error checking for simulation models
involves checking for software and input coding errors, as well as visual observations from the simulation
animation. The error check for input data included checking the connectivity of the modeled roadway
links, their number of lanes, functional classification, capacities, traffic controls at the intersections,
prohibited movements, and free-flow speeds. For checking the volume-demand data, the trip generation
zones were checked, in addition to the origins and destinations of the trips in the modeled transportation
network. Finally, the simulation model was run with the required network loading time (seed time) and
simulation periods and visually inspected for any errors or unexpected model output.

3.1.7 Development and calibration of the hybrid model

After a functional mesoscopic simulation model was developed for the base year, it was integrated within
the regional travel demand model to form the hybrid model used for the purposes of this study. To
facilitate the exchange of data between the two models, an interface was developed that allowed for input
and output data transfer between the two models. The interface was used to prepare the O-D matrix from
Cube in the desired format to be used as input for the simulation model. Also, the interface was used to
prepare the roadway link’s travel times and traffic volumes from the simulation model to be fed back into
the regional travel demand model.

To prepare the O-D matrix from the regional travel demand model that was used in the simulation model,
several files were coded using TP+ programming language to extract the required O-D matrices for the
AM peak, PM peak, and off peak periods in the desired format for the simulation model. As for the
feedback from the simulation model into the regional travel demand model, the travel time over the
roadway links from the mesoscopic simulation were prepared in a format that is compatible with Cube
software. The travel times were assigned to the network roadway links in the regional travel demand
model using several programs that were coded in TP+ programming language. That process included
skimming the link IDs from the Cube modeled network, assigning the travel time to each link, and finally
producing new travel time and distance matrices for the modeled network. The general structure of the
hybrid model is shown in Figure 3.4.

34
Figure 3.4 The structure of the hybrid model

35
The feedback from the mesoscopic simulation model into the regional travel demand model was in the
form of new travel times over the roadway links based on the output from the simulation. The regional
travel demand model used the Bureau of Public Roads (BPR) volume delay functions to estimate the
congested travel times over the links. The new travel times are different than those estimated in the
regional travel demand model based on static volume-delay functions.

After the new travel times were fed back into the regional travel demand model, the distribution of the
trips between O-D pairs was performed based on the new travel times, resulting in an O-D matrix. The
new O-D matrix was fed back into the simulation model and a new model run was performed, where the
new travel times from that run were fed back into the regional travel demand model again. The process of
feeding the new O-D matrices into the simulation model, and the new roadway travel times into the
regional travel demand model, was repeated until convergence in the roadway travel times was achieved.
After convergence in the travel times between the hybrid model runs was achieved, the model was
checked for errors where the average travel times and traffic volumes assigned for roadways were
checked against observed values. To check for convergence in the roadway travel times between different
iterations, Equation 3.3 was used.

∑ ∑
L T
(ΣQil ,t − ΣQil−,t1 ) 2
l =1 t =1
<∈ Equation 3.3
N l Nt

Where: ΣQi = Equilibrium travel time over the link,

l ,t

ΣQil−,t1 = Equilibrium travel time over the link from the previous iteration,
N lt = Number of links,
Nt = Number of time intervals, and
∈ = Critical value to reach convergence.

3.1.8 Emergency evacuation scenario modeling

Following the development and calibration of the hybrid model to the base case conditions, hypothetical
emergency scenarios were incorporated within the hybrid model. The transportation system conditions
during regional evacuations are complex and different than daily traffic conditions. Hence, modeling
hypothetical emergency scenarios requires the collection of travel demand data that is different than what
is used for modeling trips made to participate in daily activities. Data were collected for each modeled
emergency evacuation scenario to estimate number of evacuating trips and the origins and destinations of
those trips. A detailed description of the data collection and analysis efforts are provided in the next
section.

To incorporate the hypothetical emergency evacuation scenario within the hybrid model, the necessary
revisions to the transportation network were made. Also, the input O-D matrix and travel demand data
were modified for each scenario to reflect the population response to evacuation orders. In addition,
traffic control parameters were changed to reflect any traffic management plans that were used for a
particular scenario if applicable. The process of incorporating the hypothetical emergency evacuation
scenario within the hybrid model is illustrated using the case study.

The evacuation trip production and attractions rates were used as input into the regional travel demand
component of the hybrid model to obtain the O-D matrix representing the evacuation trips between
different zonal pairs. The O-D matrix, in addition to the trip loading rates developed for the evacuation
scenario, were used as input for the simulation component of the hybrid model. After that, selected

36
system performance parameters were calculated based on output from the hybrid model runs to compare
the different scenarios and traffic management options if available. The system performance data
collected consisted of evacuation time estimates, average travel times, and average travel speeds for the
different modeled evacuation scenarios. Figure 3.5 illustrates the process used to model emergency
evacuations using the developed hybrid model.

FM-COG
Data

Socio-
Network Signal
Economic
Data Timing Data
Data

Evacuation
Trip
Trip
Generation
Productions

Evacuation Travel
Trip Survey
Attractions Data

Trip
Distribution Evacuation
Trip
Loading
Rates

O-D Output

Traffic
Dynasmart-P Management
Plan

Dynamic
Traffic
Assignement

Congestion Travel
ETE
levels Speeds

Figure 3.5 Evacuation modeling methodology

37
3.2 Human Behavior Data Collection
It is vital to identify the different parameters influencing the travel decisions of individuals during
emergency evacuations to model the changes in the travel demand over the transportation system. The
reliability and accuracy of the output from evacuation models rely on the ability to model the actual
evacuee’s behavior, and their response rates to evacuation orders. In addition, human behavior data were
collected to be used as input for the hybrid model. The data were analyzed and used to develop trip
generation rates and travel demand levels, in addition to evacuee’s response rates that were not available
in the literature and were needed for modeling the evacuation scenario.

This study will evaluate two modeling scenarios. The first scenario will evacuate the Fargodome parking
lots and act as an initial test of the hybrid model. The second scenario, which is the primary scenario of
this study, will evaluate a regional evacuation triggered by river flooding. After consulting with the
stakeholders and to link the modeling efforts to possible local threats, it was decided to model the regional
evacuation event triggered by the flooding of the Red River of the North. The scenario represented
evacuating the area within the 100-year flood plain in the Fargo-Moorhead metropolitan area triggered by
a levee breach or overflow.

3.2.1 Data collection

Faced by the lack of data and performance measures related to transportation security, there was a need to
develop accurate input data to improve the accuracy and reliability of the model output. To generate the
needed data, a public survey was administered by mail where respondents were asked to mail their
responses back in prepaid envelopes (mail-in/mail-back method). The survey was a stated response
survey category, which was discussed in detail in Section 2.

3.2.2 Survey population and sampling

The surveyed population included the households located within the 100-year flood plain in the Fargo-
Moorhead metropolitan area. After the evacuation area boundaries were identified, the mailing addresses
for the households located in that area were provided by the city of Fargo, ND, and the city of Moorhead,
MN. Of the approximately 57,000 households in the Fargo-Moorhead metro area (2005), 20,613 were
located within the evacuation area, which represents 36% of the households. After the survey population
was identified, the sample size was determined using Equation 3.4 (19). Based on Equation 3.4 and using
a 95% level of confidence, the needed sample size was determined to be 377 responses. There were 1,500
surveys sent in October 2008, and 454 responses were returned; out of which, 437 provided usable data
for the purposes of this study.

𝑆 = [𝑃(1 − 𝑃)/[(𝐴2 /𝑍 2 ) + (𝑃(1 − 𝑃)/𝑁)] Equation 3.4

Where: S = Sample sSize,

N = Population size,
P = Estimate of the percentage of people in population interested (50% was used as a
conservative estimate,
A = Level of accuracy (0.05 was used), and
Z = Number of standard deviations of the sampling distribution.

3.2.3 Survey design and administration

At the design stage, the survey method, the associated time frame, and budget requirements were
established. The design of the survey has a significant effect on the response rates and accuracy of the

38
data obtained. The survey was designed to be as short as possible, keeping the directions and questions
simple, clear, objective, and complete. To increase the response rates, the survey included a cover letter
to explain the purpose and benefits of the survey, mentioned the partnership with both Cass and Clay
County Emergency Management Centers, and a respondent friendly questionnaire and map (note
Appendix A).

The final survey used consisted of a two-page (eight questions) questionnaire and a map showing the
evacuation area and possible evacuation destinations. The survey questions were designed to provide the
following data:
1) Households that will comply with evacuation orders
2) The estimated time needed to prepare for evacuation
3) Number of vehicles used in the evacuation process
4) Evacuation destination category
5) Number of zone that represents their evacuation destination
6) Household category

The data provided by the respondent were used for the following purposes:
1) Trip generation:
a) % of single-family households that will evacuate
b) % of multi-family households that will evacuate
c) Average number of vehicles used in the evacuation trip for each household category
2) Trip distribution:
a) % of single-family household that will evacuate to each zone
b) % of multi-family household that will evacuate to each zone
3) Trip loading rates:
a) % trips evacuating at each time step (mobilization time)
4) General questions:
a) Does the category of the household affect the probability of complying with evacuation orders?
b) Does the category of the household affect the time needed for evacuation?
c) Does the number of vehicles used in the evacuation affect the time needed for evacuation?

3.2.4 Data analysis

After the survey design was finished, the survey was administered and data collection from the population
sample began. While the key survey results are located in the following section, Appendix B contains the
complete survey results. The results from the survey responses were coded into the system, and the data
were tested and cleaned to keep the usable data. Afterward, the data were processed and prepared to be
analyzed. The responses were compiled in a Microsoft Office Access database using a form prepared in
Visual Basic. The Visual Basic form was designed to reduce the possibility of human errors while
compiling the data from the survey.

3.2.4.1 Trip generation: To estimate the demand for travel over the transportation system during
emergency evacuation events, the percentage of households evacuating the area and the number of
vehicles used by each household were analyzed. For the purposes of this study, special facility population
trips were not included in the simulation due to their minimal effect on the performance of the system,
where they are evacuated early during the event by air or special transportation. The households were
categorized by their size as being single-family or multi-family households. Tables 3.2 and 3.3 summarize
the data related to trip generation obtained from the sample response.

39
Table 3.2 Households Compliance with Evacuation Orders
# Single-Family HH Comply 389 # Single-Family HH Not Comply 25
# Multi-Family HH Comply 23 # Multi-Family HH Not Comply 0
# Total HH Comply 412 # Total HH Will Not comply 25

Table 3.3 Compliance Rates by Household Category

Household Type % Comply % Not Comply Total
Single-Family HH 94.0% 6.0% 100.0%
Multi-Family HH 100.0% 0.0% 100.0%

From Tables 3.2 and 3.3, most of the respondent (94.3%) stated that they will comply with evacuation
orders when issued. However, people who live in houses were more reluctant to leave than those living in
apartments. The rationale is that some of the single-family household’s residents reported they want to
stay and protect the property, and minimize the damage as much as possible.

After determining the percentage of households that will comply with the evacuation orders, the number
of vehicles used for evacuation per household was determined. From the sample obtained, it was found
that the average number of vehicles used in the evacuation was 1.62 vehicles/household. Tables 3.4 and
3.5 summarize the data related to the number of vehicles used during the evacuation.

Table 3.4 Number of Vehicles Used per Household for Evacuation

# of Vehicles Used for Evacuation Number of Households % Total
0 0 0.0%
1 189 45.9%
2 192 46.6%
>=3 31 7.5%
Total 412 100.0%

Table 3.5 Number of Vehicles Used for Evacuation by Household Category

# of Vehicles # of Single-Family % of Single-Family # of Multi-Family % of Multi-Family
Used Households Households Households Households
0 0 0.0% 0 0.0%
1 171 44.0% 18 78.3%
2 187 48.0% 5 21.7%
>=3 31 8.0% 0 0.0%
Total 389 100.0% 23 100.0%
Average/HH 1.64 Vehicles/HH 1.22 Vehicles/HH

The data obtained from the survey were classified as categorical where the conditions of multinomial
experiments were satisfied (26). Hence, the suitable method for analyzing the data was by using one-way
and two-way table’s analysis. To check if the household category affected the average number of vehicles
used during emergency evacuations, the Chi-Square test was used. The first test was to check if the
proportions of people using a certain number of vehicles were significantly different. After that, the test
was performed to check if the proportions of people using a certain number of vehicles were statistically

40
significantly different from one household category to the other. The Chi-Square test was performed at a
confidence level of 95% (α=0.05) with the following assumptions:

Ho: P1=P2=P3
Ha: At least one of them is different

Test statistic: χ2 = ∑ [ni – Ei]2 / Ei Equation 3.5

Rejection region: χ2 > χ2(0.05) with (k-1) degrees of freedom

Where: Ho: The null hypothesis (the percentages of people using a certain number of vehicles for
evacuation are the same),
Ha: The alternative hypothesis (at least one of the percentages is different),
Pi: Represents the hypothesized values for multinomial probabilities (percentage using a certain
number vehicles for evacuation),
Ei = nPi and it represents the expected cell count,
n : The total sample size, and
k: The number of categories.

From Table 3.5:

χ2 = 123.529
Degrees of Freedom (D.F.) = 3-1 = 2

From the Chi-Square tables, with α=0.05 and D.F.=2, χ2 = 5.99147 (within the rejection region). The null
hypothesis was rejected based on the data provided in Table 3.6. Hence, it was concluded that at least one
of the proportions of the number of vehicles used during emergency evacuations was significantly
different than the others.

Table 3.6 Chi-Square Goodness-of-Fit Test for Categorical Variable

Category Observed Test Proportion Expected Contribution to Chi-Sq
1 189 0.333333 137.333 19.4377
2 192 0.333333 137.333 21.7605
3 31 0.333333 137.333 82.3309
N N* DF Chi-Sq P-Value
412 0 2 123.529 0

To test if the number of vehicles used for the evacuation was independent from the type of the household,
the two-way (contingency) table analysis was used. The test was performed at α=0.05 (95% Confidence
Interval) with the following assumptions:

Ho: The two classifications are independent

Ha: The two classifications are dependent

Test statistic: χ2 = ∑ [nij – Eij]2 / Eij Equation 3.6

Rejection region: χ2 > χ2(0.05) with [(r-1)*(c-1)] degrees of freedom

Where: Eij = (RiCj)/n,

r : The numbr of categories of the first classification, and
c : The number of categories of the second classification

41
At α=0.05 and D.F.=2, χ2 = 5.99147 (within the rejection region). The null hypothesis was rejected based
on the data summarized in Table 3.6. Hence, the conclusion based on the Chi-Square test was that
different household categories use different average number of vehicles for evacuations.

Table 3.7 Chi-Square Test for Number of Cars Used vs. Household Type
Cars
1 2 3 Total
HH
171 187 31
1 178.45 181.28 29.27 389
0.311 0.18 0.102
18 5 0
2 10.55 10.72 1.73 23
5.259 3.051 1.731
Total 189 192 31 412

In Table 3.7, the expected counts were printed below observed counts and Chi-Square contributions were
printed below expected counts. The χ2 value was 10.634 with two degrees of freedom, while the P-Value
was (0.005), and one cell with an expected count less than 5. Based on the survey results, Equation 3.7
was used for the evacuation trip generation.

Number of trips = 1.64*(evacuating single-family HH) + 1.22 * (evacuating multi-family HH)

Equation 3.7

3.2.4.2 Trip distribution: After the demand for travel during the emergency event was estimated, the
destinations for evacuation trips were determined based on data from the survey responses. The modeled
area was divided into 15 possible internal destinations and four destinations outside the metropolitan area.
The survey respondents were asked to choose the zone they would most likely evacuate to if the
authorities issued the evacuation order for a flood event. Figure 3.6 shows the different possible
evacuation destinations.

42
Figure 3.6 Evacuation destinations for a flood event

Some of the evacuation zones were chosen by less than five respondents, however, for statistical analysis
purposes at least five respondents need to pick a certain zone. To eliminate zones that have less than five
responses, the responses from those zones were added with neighboring zones. Table 3.8 summarizes the
survey respondent’s choice of evacuation destinations.

Table 3.8 Households Choice of Evacuation Destination (Modified)

Zone Number 1 2+3 4 5 6+9
Frequency 37 6 27 25 12
Percentage 8.98% 1.46% 6.55% 6.07% 2.91%
Zone Number 7 8 10 11+13+15 12
Frequency 15 12 7 6 11
Percentage 3.64% 2.91% 1.70% 1.46% 2.67%
Zone Number 14 16 17 18 19
Frequency 17 52 48 29 108
Percentage 4.13% 12.62% 11.65% 7.04% 26.21%

43
The data were tested to check if the proportions of the population evacuating to a certain destination were
statistically significantly different from one zone to another. Since the data were classified as categorical
and the conditions of multinomial experiments applied to it, a one-way table method was used to analyze
the data. The test was performed at α=0.05 (95% Confidence Interval) using Equation 3.5. Table 3.9
provides a summary of the results of the Chi-Square test used for this analysis. The hypothesis used for
the Chi-Square test includes the following:

Ho: P1=P2=P3= ……. = P15

Ha: At least one of them is different

From Table 3.8:

χ2 = 363.631
D.F. = 14

From the Chi-Square tables, with α=0.05 and D.F.=14, χ2 = 23.6848 (within the rejection region).
The test showed that the null hypothesis was rejected, and that at least one of the proportions was
significantly different than the others.

Table 3.9 Chi-Square Goodness-of-Fit Test Destination Choice

Category Observed Test Proportion Expected Contribution to Chi-Sq
1 37 0.0666667 27.4667 3.309
2 6 0.0666667 27.4667 16.777
4 27 0.0666667 27.4667 0.008
5 25 0.0666667 27.4667 0.222
6 8 0.0666667 27.4667 13.797
7 15 0.0666667 27.4667 5.658
8 12 0.0666667 27.4667 8.709
10 7 0.0666667 27.4667 15.251
11 6 0.0666667 27.4667 16.777
12 15 0.0666667 27.4667 5.658
14 17 0.0666667 27.4667 3.989
16 52 0.0666667 27.4667 21.913
17 48 0.0666667 27.4667 15.35
18 29 0.0666667 27.4667 0.086
19 108 0.0666667 27.4667 236.127
N N* DF Chi-Sq P-Value
412 0 14 363.631 0

Since most of the cells have less than five counts in the “multi-family” household category, no Chi-
Square test was performed to check if the category of the household affected the choice of destination
zones in cases of regional emergency evacuations. Hence, the percentage of trips made to each destination
was used for the evacuation trip distribution. Table 3.10 summarizes the results for the trip distribution
step.

44
Table 3.10 Percentages Evacuating to Each Destination by Household Category
Zone 1 2+3 4 5 6+9
Single-Family HH 9.25% 1.29% 6.43% 6.43% 2.83%
Multi-Family HH 4.35% 4.35% 8.70% 0.00% 4.35%
Zone 7 8 10 11+13+15 12
Single-Family HH 3.60% 3.08% 1.54% 1.29% 2.57%
Multi-Family HH 4.35% 0.00% 4.35% 4.35% 4.35%
Zone 14 16 17 18 19
Single-Family HH 4.11% 12.08% 12.08% 6.94% 26.48%
Multi-Family HH 4.35% 21.74% 4.35% 8.70% 21.74%

From the survey responses, 57.52% of the total evacuees would leave the metro area. Although the
destinations differ between household category, 56.53% of the multi-family evacuees and 57.58% of the
single-family evacuees would leave the Fargo-Moorhead area.

3.2.4.3 Trip loading rates: Due to the dynamic nature of the emergency evacuation problem, the trip
loading rates are very important to achieve realistic representation of the traffic conditions in the modeled
area. In the hybrid model, traffic data were examined for 14 major intersections in the modeled area to
determine the trip loading rates during daily peak hours. For evacuation trips loading rates, respondents
were asked to provide estimates of their mobilization, which is shown in Table 3.11.

Table 3.11 Time Needed to Prepare of Evacuation

1-15 min. 16-30 min. 31-45 min. 46-60 min >60 min.
23 88 80 112 109
5.6% 21.4% 19.4% 27.2% 26.5%

The data from the survey were compared to the estimated data using the Kolmogorov-Smirnov two-
sample test. The test statistic (T) represents the maximum difference between the two cumulative
distributions being compared to determine if they have the same distribution. The null hypothesis was that
both data sets have the same distribution, and the alternative hypothesis was that they are different. The
value of the test statistic was 0.15498 and the critical value of (T) at a 95% level of confidence was
0.2935. Hence, the null hypothesis could not be rejected since the two distributions were not statistically
different.

3.2.4.4 Data comparison: The data collection and analysis showed that there was a distinct variation in
the response and behavior of small- and medium-size urban areas from that of inhabitants of large cities.
The data collected for evacuee’s behavior provided rates that were different than the averages discussed
in Section 2, which were mainly based on data obtained from hurricane regions evacuations and surveys.
In addition, the trip loading rates function developed in this study was different than the general function
provided in the literature.

Several estimates were reported in the literature of the average number of vehicles used per household
during emergency evacuations, where an average of 1.3 vehicles was used in the literature (23). Ruch and
Schumann (1997) estimated the number of evacuating vehicles to be 1.35 based on a behavioral survey,
while Prater et al. (2000) reported a rate of 1.34 vehicles per HH for Hurricane Bret (28). The data

45
obtained from the survey used in this study showed that an average of 1.62 vehicles per household was
estimated to be used in cases of regional emergencies. Also, the survey data showed that multi-family
households reported an average of 1.22 vehicles per household, while single-family households reported
an average of 1.64 vehicles per household.

The evacuation trips loading rates are determined by the time families need to prepare and leave the
household after receiving the evacuation orders. The evacuation trip loading rates have a significant effect
on the performance of the transportation system performance during emergency evacuations. Figure 3.7
shows how the curve representing the fitted data from this study compared with the data developed by
Tweedie et al. (5). It is shown from the figure that evacuation trips are loaded at a higher rate using the
function developed by Tweedie et al. than the function developed from the survey data. The evacuation
trip loading rates affect the congestion levels and the clearance time needed by evacuees during
emergency evacuations.

Trip Loading Curves

1.2

1
Percenatge Evacuating

0.8

0.6
Fitted Data
Tweedy et al
0.4

0.2

0
0 20 40 60 80 100 120 140 160
Mobilization Time (minutes)

Figure 3.7 Mobilization time curves

46
4. CASE STUDY AND MODELED SCENARIOS
This section provides a description of the case studies used for this study. It illustrates how the modeling
approach was used to develop different emergency evacuation scenarios, and it summarizes the case study
results. In addition, this section provides a discussion of the study results and the conclusions based on
those results.

4.1 Case Study

For the purpose of this study, the 2005 regional travel model for the Fargo-Moorhead (F-M) area was
used. The F-M model consists of 568 Traffic Analysis Zones (TAZs) and a network of 1,709 nodes that
are connected by 2,412 links. The modeled area includes four jurisdictions: the cities of Fargo and West
Fargo in North Dakota, as well as Moorhead and Dilworth in Minnesota. The total population for the
metropolitan area was estimated at 190,500 in 2005 with a projected annual growth rate of 1% over the
next 30 years (16). Figure 4.1 shows the modeled area.

The F-M travel demand model was chosen for this study because it provided the adequate size and level
of detail that is sufficient for the purpose of this study. In addition, the availability of data, which were
provided either through the Fargo-Moorhead Council of Governments (F-M COG) or through the
Advanced Traffic Analysis Center (ATAC), played a role in selecting the model.

This study explored the use of DYNASMART-P, which is recognized by the FHWA for evacuation
applications in conjunction with the traditional four-step travel demand model. The value of this approach
is in utilizing the available data and modeling resources to develop evacuation models that are cost
efficient, relatively easy to develop and update, and capable of capturing traffic and drivers’ behavior
under different traffic management measures.

47
Figure 4.1 The modeled Fargo-Moorhead metropolitan area

4.2 Fargo-Moorhead Regional Travel Demand Model

Several steps were involved in the process of constructing and calibrating the F-M COG’s regional travel
demand model; those steps included data preparation, trip generation, trip distribution, mode split, traffic
assignment, and model calibration. A detailed description of those steps was provided in Section 3. The
input data needed for the development and calibration of the regional travel demand model was either
provided by the F-M COG or generated from review of the available literature and primary data collection

48
efforts. The F-M COG regional travel demand model was developed using the TP+ modeling system
produced by Citilabs within their Cube software system.

4.2.1 Trip generation

Trip generation models estimate the number of trips generated for each TAZ in the modeled metropolitan
area using the social and economic attributes for that TAZ. Trip productions were estimated using factors
such as the number of households in that area, household sizes, and age groups within the household. Trip
attractions were estimated using variables such as employment levels and commercial floor space. Trip
generation models utilized the zonal and external trip data as input and generated an array of production
and attraction. The values within the array represented the number of trips produced within and attracted
to each internal TAZ or to external TAZs.

4.2.1.1 Internal zones trip productions: The number of trips produced by each TAZ was based on the
number of single-family and multifamily households in that TAZ. Trip productions were estimated
through multiplying the total number of single-family or multifamily dwelling units by the appropriate
daily vehicle trip rates. The trips were separated into HBW, HBO, and NHB production trips by
multiplying the total vehicle-trips with the percentage of trips by purpose. The vehicle trip rates were
adjusted during the calibration process. Table 4.1 is based on NCHRP 365 and it summarizes the trip
generation rates used in the F-M COG regional travel demand model.

Table 4.1 Vehicle Trip Generation Rates

Percentage of Trips by Purpose
Dwelling Daily Vehicle
HBW HBO NHB
Category Trip Rate
Single-Family 9.55 0.20 0.57 0.23

Multi-family 6.47 0.20 0.57 0.23

4.2.1.2. Internal zones trip attractions: To estimate the number of trips attracted to each zone, the
TAZs representing the metropolitan area were classified as being within the central business district
(CBD) area or within the non-central business district (NCBD). Different trip attraction rates were used
for HBW, HBO, and NHB trip attractions for CBD and NCBD zones. Table 4.2 summarizes the trip
attraction rates used in the F-M COG regional travel demand model, which were based on NCHRP 365.

Table 4.2 Trip Attraction Rates

Trip Purpose CBD Zones NCBD Zones
HBW 1.45 x TE 1.45 x TE
HBO 2.0 RE + 1.7 SE + 0.5 OE + 0.9 HH 9.0 RE + 1.7 SE + 0.5 OE + 0.9 HH
NHB 1.4 RE + 1.2 SE + 0.5 OE + 0.5 HH 4.1 RE + 1.2 SE + 0.5 OE + 0.5 HH

Where: TE = Total Employment,

RE = Retail Employment,
SE = Service Employment,
OE = Other Employment, and
HH = Households.

49
 4.2.1.3 University trip productions and attractions: Due to the special nature of trip making behavior
for university trips, North Dakota State University (NDSU), Concordia College, and Minnesota State
University Moorhead (MSUM) were treated as special trip generators. A home based university trip
category was included in the F-M COG regional travel demand model. To estimate the number of trips
generated by the college campuses and the area directly affected by them, NDSU was used as a model.
Data related to trips made from and to NDSU were gathered by ATAC through interviews and parking
data, and it was determined that the number of college trips could be predicted based on variables that can
be forecasted by the F-M COG. Table 4.3 provides the university trip rates used in the model (1).

Table 4.3 University Trips Generation

Predicted 2005 Enrollment
Concordia
Purpose Rate Population Category MSUM NDSU
College
HBW
0.16 On-Campus Students 1,794 1,559 2,876
Productions
HBO
0.37 On-Campus Students 1,794 1,559 2,876
Productions
NHB
0.17 Total Students 2,608 7,491 11,723
Productions
HBS
0.12 On-Campus Students 1,794 1,559 2,876
Productions
HBW
0.30 Total Students 2,608 7,491 11,723
Attractions
HBO
0.44 Total Students 2,608 7,491 11,723
Attractions
NHB
0.17 Total Students 2,608 7,491 11,723
Attractions
HBS
0.72 Off-Campus Students 814 5,932 8,847
Attractions

4.2.1.4 High school and grade school trip generation: In the F-M COG regional travel demand model,
home-based school trips were calculated independently. The initial estimate of HBS trip attractions was
set to equal the number of students enrolled in the school zone. Due to the different nature of school trips
made by students who may possess a driver’s license, the student population was divided into high school
and grade school students.

4.2.1.5 Airport trip generation: The Fargo Hector International Airport is located within the areas
modeled in the F-M COG regional travel demand model. In 2005, there were 549,209 passengers using
the airport (30). The total number of passengers was used to estimate the average daily number of
passengers using the airport. The ITE trip generation manual provided the rates used for airport trip
generation.

4.2.1.6 External trip generation: Trips with both ends outside the modeled metropolitan area are known
as external-external (EE) trips. In the F-M COG regional travel demand model, those trips are assumed to
account for 10% of the interstate traffic. If only one trip end is within the modeled area, then it is defined
as being an internal-external (IE) trip or external-internal (EI) trip. In the F-M COG regional travel
demand model, trip attractions for external zones were estimated by multiplying the average daily traffic
with the percentage of trips by purpose at each external zone. To calculate the number of productions for
the interstate highways, the total number of through trips was subtracted from the ADT and then
multiplied by the percentage of trips by purpose.

50
4.2.1.7 Trip generation adjustment: Because different factors were used to estimate trip productions
and attractions in the F-M COG regional travel demand model, the trip production and attraction totals
were unbalanced. In reality, each production must be paired with an attraction to form a trip, and the trip
production totals must match the trip attraction totals for each trip type. Equations 3.1 and 3.2 were used
to balance the total trip productions and attractions, and the balanced results are summarized in Table 4.4
(1).

Table 4.4 Total Adjusted Productions and Attractions by Trip Purpose

Trip Purpose Total Trip Productions Total Trip Attractions
HBW 159,347 159,347
HBO 452,513 452,513
NHB 99,546 99,546
HBS-University 9,942 9,942
HBS-High School 9,027 9,027
HBS- Grade School 20,185 20,185

4.2.2 Trip distribution

The output from the trip generation step was represented by the number of trips produced by and attracted
to each TAZ in the modeled metropolitan area. In the trip distribution step, the trip ends are connected to
establish the flow of trips from production zones to attraction zones. The output from the trip distribution
is a matrix of trips produced and attracted between the TAZs called the origin-destination (O-D) matrix.

The gravity model was used to distribute trips between the TAZs in the F-M COG regional travel demand
model. In the gravity model, the number of trips is assumed to be based on the level of activity of the
zones and the distance (travel cost) between those zones. The gravity model is based on the assumption
that trip interchange between two TAZs is proportional to the number of trips generated at each TAZ and
is inversely proportional to the cost of traveling between those two TAZs. Equation 4.1 represents the
gravity model used in the F-M COG regional travel demand model (29).

𝑃𝑖 [𝐴𝑖 𝑓𝑖𝑗 𝑘𝑖𝑗 ]

𝑇𝐼𝐽 = ∑𝑚
Equation 4.1
𝑛=1 𝐴𝑗 𝑓𝑖𝑗 𝑘𝑖𝑗

Where: 𝑇𝐼𝐽 = The flow of trips from each production zone i to each attraction zone j,
𝑃𝑖 = Total number of trips produced in zone i,
𝐴𝑗 = Total number of trips attracted to zone j,
𝑓𝑖𝑗 = Friction factor, and
𝑘𝑖𝑗 = Adjustment factor for trip interchanges between zones i and j.

The friction factor in the gravity model is inversely proportional to the travel cost between zones i and j.
In the F-M COG regional travel demand model, the friction factors were adjusted until the observed and
predicted trip length distributions were similar. The k factor is a scaling factor that is used during
calibration and it limits or increases the volume of traffic that crosses sections of the network.

4.2.3 Mode split and temporal distribution

In the F-M COG regional travel demand model, automobiles are the only mode of transportation modeled
because of the low percentage of public transit use. The number of trips generated in the model is
represented by vehicle trips per day; however, the trips needed to be assigned based on hourly increments.

51
Based on several hourly traffic counts throughout the metropolitan area the daily traffic was divided into
hourly peaks, as shown in Table 4.5.

Table 4.5 Peak Hour Percentage of Daily Traffic

Period Time % of daily traffic
AM 7:45 AM - 8:45 AM 7.53%
PM 5:00 PM – 6:00 PM 8.47%
Off Peak All other times 84%

The production attraction matrix was added to the transposed production attraction matrix and then the
trips were divided by two. Using this method, it was assumed that half of the trips go from production to
attraction and half of the trips are returning from attraction back to the production. The matrix was then
multiplied by the appropriate time of day percentage to obtain three origin destination matrices: AM, PM,
and off peak. Also, factors provided by NCHRP 365 were used to produce peak hour directional volumes
for HBW, HBO, and NHB trips to allow for peak hour traffic assignment that were representative of the
peak hour direction and hourly trip percentage.

4.2.4 Traffic assignment

The last step in the F-M COG regional travel demand model was the assignment of the predicted trip
flows between origin and destination TAZs to modeled roadways in the metropolitan area. The
assignment step started with three separate O-D matrices: AM, PM, off peak, which contained the traffic
volumes to be assigned for each O-D pair. User-equilibrium traffic assignment method was used for this
model, and it was implemented using a travel cost (time) function to evaluate the most desirable path for
the trip. Travel time was set to the free flow travel time for the first iteration and then changed with
iterations depending on congestion levels on the roadway network. This iterative process continued until
there was no available path at which the cost could be reduced.

4.2.5 F-M COG regional travel demand model calibration

The final stage in the process of developing the F-M COG regional travel demand model was the
calibration and validation of the model to the 2005 base year traffic data. The goal of the calibration
process is to have the predicted travel behavior match the observed travel behavior data available. The
process of calibrating different parameters and error checking was performed at each modeling step to
minimize the errors in the overall model. The calibration process was illustrated in Figure 3.4, and a
summary of the different stages of model calibration is provided in the following sections.

4.2.5.1 Trip length distribution: The first task of the calibration of the F-M COG regional travel
demand model was to check how the modeled trip length distribution compared with the trip length
distribution provided by the census for transportation planning (CTPP) data for the modeled metropolitan
area. The modeled HBW, HBO, and NHB trip length distributions were compared with those provided by
the CTPP. If the trip length distribution of the modeled trips did not match that provided in the CTPP
data, then the friction factors were adjusted until the two distributions were similar. The friction factors
were adjusted using Equation 4.2.

𝑇𝑡𝑜𝑏𝑠
𝐹𝑡𝑖+1 = 𝐹𝑡𝑖 ∗ Equation 4.2
𝑇𝑡𝑖

Where: 𝐹𝑡𝑖+1 = The friction factor for time interval t for iteration i+1,
𝐹𝑡𝑖 = Friction factor for time interval t for iteration i,

52
 𝑇𝑡𝑜𝑏𝑠 = The observed number of trips in time interval t, and
𝑇𝑡𝑖 = The estimated number of trips in time interval t for iteration.

Several functions could be used to represent friction factors, but the gamma distribution is one of the best
distributions that could be used (32). Equation 4.3 represents the function used for the F-M COG regional
travel demand model friction factors (32).

𝑏
𝐹𝑖𝑗 = 𝑎 ∗ 𝑡𝑖𝑗 ∗ 𝑒 𝑐∗𝑡𝑖𝑗 Equation 4.3

Where: 𝐹𝑖𝑗 = The friction factor between zones i and j,

𝑎, 𝑏, 𝑎𝑛𝑑 𝑐 = Model coefficients are “b” and “c”, while “a” is a scaling factor that could be
changed without changing the distribution,
𝑡𝑖𝑗 = Travel time between zones i and j, and
𝑒= The base of the natural logarithms.

The factors initially used for the F-M COG regional travel demand model are summarized in Table 4.6,
and the friction factors functions used are shown in Figure 4.2.

Table 4.6 Friction Factors Model Coefficients for the F-M COG Model
Trip Purpose HBW HBO NHB
A 1 1 1
B -0.351 0.548 1.110
C -0.043 -0.212 -0.280

Friction Factors

1.6

1.4

1.2

1
Fij

0.8

0.6

0.4

0.2

0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
tij

HBW HBO NHB

Figure 4.2 Friction factor functions used for the F-M COG model

53
 4.2.5.2 Vehicle miles traveled: The total vehicle miles traveled (VMT) values are based on the number
of generated trips in the model and their corresponding lengths in miles. The total modeled VMT values
were first compared with the observed total VMT values for both the North Dakota and Minnesota sides
of the metropolitan area. If the observed and reported VMT values were different, then the trip generation
rates were adjusted to make them similar on a regional level.

After the total VMT values generated by the travel demand model were calibrated on a regional level, the
VMT values were checked by roadway functional class. To adjust the modeled VMT values to match the
observed VMT values, the global speeds by land use characteristics and node delays were adjusted to
make them similar. For the F-M COG regional travel demand model, the overall modeled VMT was
within 2% of the observed VMT for the modeled metropolitan area, which is within the 5% difference
limit provided in the travel model improvement program’s model validation and reasonable checking
manual (17). Table 4.7 provides a summary of the observed and modeled VMT values for the F-M COG
regional travel demand model.

Table 4.7 VMT Values by Jurisdiction for the F-M COG Model
Difference in % Difference in
Jurisdiction VMT Reported VMT Modeled
VMT VMT
Fargo 1,845,042 1,823,416 -21,626 -1.17%
Moorhead 482,413 430,514 -51,899 -10.76%
West Fargo 169,523 172,657 3,134 1.85%
Dilworth 41,029 71,825 30,796 75.06%
ND 2,014,565 1,996,073 -18,492 -0.92%
MN 523,442 502,339 -21,203 -4.03%
Metropolitan Area 2,538,007 2,498,412 -39,595 -1.56%

4.2.5.3 Screenline counts: The screenlines used for the F-M COG regional travel demand model
included I-29, I-94, the Red River, and the main railroads tracks. The total ADT values for the roadways
crossing the screenlines were compared with the modeled daily traffic volumes on those roadway links. If
the ADT values were different from the modeled traffic volumes for those links, the K factors were
adjusted to make them closer.

The K factors are trip distribution factors that are used to adjust the utility of making trips between TAZ
pairs; they are used in the gravity model to match the modeled trip interchange between the TAZs with
the observed trip distribution in the modeled metropolitan area. If the modeled traffic volumes crossing
the screenline were above the specified criteria, a lower k factor was assigned to inhibit trips crossing the
screenline. Similarly, if the modeled traffic volumes crossing the screenline had a modeled traffic volume
below the designated criteria, a higher k factor would be applied to affected zones, making trips between
zonal pairs crossing the screenline more attractive.

After an accurate screenline distribution was achieved for the modeled area, the calibration process was
repeated starting with checking the trip length distribution until all the successive calibration components
were within criteria. Table 4.8 provides a summary of the K factors used in the F-M COG regional travel
demand model and the traffic volumes crossing the screenlines in the modeled metropolitan area.

54
Table 4.8 Screenline Counts and K Factors for the F-M COG Model
Modeled Traffic Volume Percent
Screenline K Factor ADT
ADT Difference Difference (%)
Interstate 29 0.80 96,200 91,500 -4,700 -4.89%
Interstate 94 0.33 135,075 136,400 1,325 0.98%
Red River 0.30 109,950 110,600 650 0.59%
Railroad 0.40 122,875 122,800 -75 -0.06%

4.2.5.3 Network wide adjustments: The final step in the regional travel demand model calibration was
to check how the modeled traffic volumes on the roadway links compared with the ADT values obtained
from field counts in the metropolitan area. If the ADT values and modeled traffic volumes were
significantly different, then the global speeds were adjusted to make them similar. Table 4.9 summarizes
the results of modeled and observed traffic volumes comparison for the F-M COG regional travel demand
model.

Table 4.9 Traffic Volume Comparison for the F-M COG Model
Volume Range Above Criteria Meets Criteria Below Criteria % Within Criteria
AADT>25,000 0 18 1 95%
25,000 to 10,000 6 131 23 82%
10,000 to 5,000 35 134 22 71%
5,000 to 2,500 33 129 15 72%
2,500 to 1,000 46 72 13 56%
AADT<1,000 34 27 2 43%
Total 154 511 76 69%

To test for the overall difference between ADT values and modeled traffic volumes, the root mean square
error (RMSE) measure was used. The RMSE value was found by averaging the square error for the traffic
volume for each link and then taking the square root for the averages. Table 4.10 summarizes the RMSE
values for the modeled roadway links by traffic volume range (17).

Table 4.10 RMSE Value by Volume Range for the F-M COG Model
Volume Range RMSE (%) Typical Limits (%)
AADT>25,000 14 % 15-20 %
25,000 to 10,000 24 % 25-30 %
10,000 to 5,000 37 % 35-45 %
5,000 to 2,500 55 % 45-100 %
2,500 to 1,000 93 % 45-100 %
AADT<1,000 >100 % >100 %

4.3 Fargo-Moorhead Hybrid Model

Two types of input files are used for the DYNASMART-P software: simulation input files and graphical
input files. The needed files for DYNASMART-P were prepared using the DYNASMART-P editor
(DSPED), Microsoft Excel, and Cube software. These files are text based, and each one of them must be
prepared in a certain format provided in the user manual (20).

55
An interface between Cube and DYNASMART-P software was created. Using this interface, the static O-
D matrix from Cube was prepared in the desired format for DYNASMART-P to be used as input. Also,
the interface was used to feed the output link traffic volumes from the DYNASMART-P simulation runs
into the travel demand model in Cube. Figure 4.3 shows the Fargo-Moorhead hybrid model in
DYNASMART-P.

Figure 4.3 Screenshot from the Fargo-Moorhead DYNASMART-P model

4.3.1 Fargo-Moorhead hybrid model calibration

The feedback from DYNASMART-P into the Cube model was represented by new travel time over the
links based on the dynamic traffic assignment procedures. The original O-D input matrix for
DYNASMART-P was based on the calibrated F-M COG travel demand model. For the hybrid model,
traffic data were examined from 14 signalized intersections in Fargo, ND, to determine how the trips are
loaded onto the network during daily peak hours (45). These 14 intersections collect continuous traffic
volume data and volume variation throughout the day was analyzed for those intersections to determine
traffic loading rates during the daily peaks. The traffic volumes were averaged for those intersections
during daily traffic peaks, and these averages were used to determine the trip loading patterns into the
mesoscopic simulation software. Figure 4.4 shows the locations of the intersections used to study the
traffic patterns in the metropolitan area. After the initial simulation runs, the O-D matrix loading rates
were adjusted to reflect the actual traffic conditions on the ground.

56
Figure 4.4 Intersection locations used to collect traffic volume data (Fargo, ND)

After providing adequate seed time (traffic loading) and simulation periods for the hybrid model, several
runs were conducted until convergence in link travel times was achieved using Equation 4.4 where the
value obtained was 0.0217 (42).

𝑙,𝑡 𝑙,𝑡 2
𝐿 𝑇
�∑𝑙=1 ∑𝑡=1(∑ 𝑄𝑖 −∑ 𝑄𝑖−1 )
<𝜖 Equation 4.4
𝑁𝑙 𝑁𝑡

Where, ΣQi = Equilibrium travel time over the link,

l ,t

ΣQil−,t1 = Equilibrium travel time over the link from the previous iteration,
N lt = Number of links,
N t = Number of time intervals, and
∈ = Critical value to reach convergence (0.05).

For model validation purposes, the overall model was examined in terms of the acceptable levels of error,
average travel times, and average trip lengths, it was found to be within the criteria. Table 4.11
summarizes the traffic assignment by volume range from the calibrated hybrid model.

57
Table 4.11 Model Assignment by Traffic Volume Range
Volume Range % Within Criteria Criteria % Deviation
ADT > 25,000 94% ± 22%
25,000 to 10,000 92% ± 25%
10,000 to 5,000 82% ± 29%
5,000 to 2,500 75% ± 36%
2,500 to 1,000 60% ± 47%
ADT < 1,000 59% ± 60 %

The results from the mesoscopic simulation runs will vary due to the difference in the random seed
number used for each model run. To determine the number of simulation runs required, the variance of
the traffic volumes on roadway links from simulation results needed to be determined. A number of
simulation runs were performed to calculate the mean and standard deviation of the traffic volumes of the
simulation output. Equation 4.5 was used to determine the number of simulation runs needed (24).

𝛿
𝑁 = [𝑡𝛼/2 � �]2 Equation 4.5
𝜇∗𝜀

Where: N = Number of simulation runs required,

tα/2 = The critical value of the t-distribution at confidence level (1-α),
δ = Standard deviation of the performance measure based on simulation,
µ = Mean value of the performance measure based on simulation, and
ε = Allowable error (5%).

For the purposes of this study, a confidence level of 95% was used. The critical value of the t-distribution
was (2.77645) and the value of mean was 234.4 vehicles. After applying Equation 4.5, the number of
simulation runs required was four model runs.

4.4 Evacuation Scenarios

Several emergency evacuation scenarios were modeled in this study, which were developed in
consultation with local emergency managers and interested stakeholders. The methodology was initially
tested using a hypothetical evacuation scenario of the Fargodome. After that, scenarios were modeled for
evacuation events triggered by flooding of the Red River. For each scenario, the location, threat level,
warning times, and the affected area’s shape and boundaries were analyzed and determined.

4.4.1 FARGODOME evacuation scenario development

To test the developed methodology, a hypothetical emergency evacuation of the Fargodome was initially
modeled. The hypothesized scenario for this study involved the evacuation of a crowd attending a
weekday evening event in the Fargodome, which is located north of the downtown business district. The
average attendance of a large scale event at the Fargodome is 21,000 people.

The Fargodome has 3,790 parking stalls in the parking lots, and it is estimated that 3,000 cars park in the
surrounding neighborhoods and parking lots. The modeled scenario measured the evacuation time
estimate for evacuating the Fargodome population to the nearest point out of the danger area and added
six new evacuation origins representing the parking lots (shown in red in Figure 4.5) and parking areas
surrounding the FARGODOME. It should be pointed out that the ETE does not include the time required
to exit the Fargodome building and to walk to the appropriate vehicle. Also, it was assumed that traffic

58
heading into the Fargodome area would be blocked and sent to other areas. In DYNASMART-P, the
vehicles generated from the different zones are loaded directly on the roadway network, and it is possible
to control the distribution of traffic into the different roadway links. Nevertheless, DYNASMART-P is
unable to model the effect of traffic conflict at the exit points of the parking lots on the capacity of those
points. The evacuation time estimate for this scenario was 34 minutes.

Figure 4.5 Fargodome evacuation DYNASMART-P links/nodes

4.4.2 Emergency evacuation scenario development

The Fargo-Moorhead metropolitan area includes the cities of Fargo, West Fargo, Moorhead, and
Dilworth. In 2005, the metropolitan area had a population of 190,500 (16), thus it was classified as a
medium size MPO (metropolitan area with a population of 50,000 to 200,000).

The Fargo-Moorhead area can be threatened by flooding from three rivers: Sheyenne River, Wild Rice
River, and Red River. The Sheyenne River diversion, which was completed in 1992, provides flood
protection for West Fargo, ND. The Wild Rice River is a tributary of the Red River (as is the Sheyenne
River) and flows into the Red River just south of the metropolitan area. In the past 112 years, it was
reported that the Red River water levels reached flood stage levels 56 times in the Fargo-Moorhead area,
exceeding the major flood stage (30 ft) 20 times (46). The Fargo-Moorhead area is the last major
metropolitan area on the Red River that remains unprotected from the 100-year (38.2 ft) flood levels (46).
In 2009, the Red River had a record crest of 40.84 ft. Due to the extended wet periods over the years, the
100-year flood plain is being revised for Fargo-Moorhead area.

There was interest from the F-M COG and other stakeholders in introducing the security aspect into
transportation planning. In the past, the main focus of key stakeholders (i.e., emergency management, law
enforcement, fire departments, and transportation agencies) was on general disaster planning and mock
drills. However, evacuation modeling would facilitate effective use of the transportation system assets to
support emergency management functions, including expedited evacuation and recovery efforts.

The key stakeholders realized the importance of analyzing the redundancies of the transportation network
associated with moving a large number of people during emergency evacuations. After consulting with

59
the stakeholders and to link the modeling efforts to possible local threats, it was decided to model the
regional evacuation event triggered by the flooding of the Red River. The scenario represents evacuating
the area within the 100-year flood plain triggered by a levee breach or overflow. Figure 4.6 shows the
metropolitan area and areas within the 100-year flood plain area.

Figure 4.6 The 100-year flood level in the Fargo-Moorhead area (38.2 ft)

60
4.4.3 Emergency evacuation scenario modeling

Three different evacuation scenarios were modeled: a possible levee breach on the Minnesota side, a
possible levee breach on the North Dakota side, and a possible levee breach or overflow on both sides of
the river. A fourth scenario was also modeled to measure the effectiveness of modifying the traffic
controls to facilitate the evacuation process. In addition, the possible levee breach or overflow on both
sides of the river was modeled using data based on the national averages to test how local data are
different than national data. In all of the modeled scenarios, it was assumed that the emergency event
occurs during the PM peak hour, and that a portion of the population (those who were at work) is allowed
to return to their homes before they start the evacuation trip. As the Red River rises during a flood event,
several roadways/intersection are closed due to flooding or dike construction. Table 4.12 lists some of the
transportation network changes based on the river stage. The closures at the 100-year flood stage were
coded into the hybrid model for the five flood evacuation scenarios.

Table 4.12 Traffic Network Changes at Different Flood Levels

River
Street Location Responsibility Reason
Stage (ft)
18' Elm St 14th - 15th Ave N Fargo Flooding
24' N. Broadway North of 37th Ave Fargo/Clay Co. Flooding
9th Ave/N.
25' Mickelson Field Park Board/Fargo Flooding
River Rd
28' 12th Ave N. Red River Bridge Fargo/Moorhead/Bridge Company Flooding
30' County Road 20 10th St - Broadway Cass Co./Clay Co. Flooding
30' 2nd St N. 1st - 5th Ave Fargo Dike
31' 2nd St S. Main Ave - 4th St Fargo Dike
33' Lower Terrace At Elm Street Fargo Flooding
34' Oak St 8th - 11th Ave N Fargo Dike
34' 1 Ave N. Red River Bridge Fargo/Moorhead Dike
35' South Terrace 125 South Terrace Fargo Dike
36' 15th Ave N. At Elm Street Fargo Dike
36' 14th Ave N. Oak - Elm Street Fargo Dike
37' Lindenwood Dr 3rd - 4th St Fargo Dike
37' S University Dr 40th - El Cano Dr Fargo Dike

4.3.3.1 Minnesota side evacuation scenario: The input for the hybrid model consisted of the modified
street network, the evacuation traffic demand, and the evacuation O-D matrix, in addition to the
associated trip loading rates. The output from the simulation consists of different system-wide averages in
addition to different performance measures on the traffic network. Table 4.2 summarizes the changes to
the traffic network during flooding conditions.

For this scenario, the evacuation population was represented by the people living within the 100-year
flood plain on the Minnesota side. The trip generation rates and compliance rates developed in Section 3
were used to obtain the number of vehicles to be loaded onto the traffic network. The evacuation O-D
matrix was loaded onto the simulation model using the function described by Equation 3.6. Based on the
evacuation area, trip generation, evacuation compliance, and evacuation destination, it would take 180
minutes (3 hours) to evacuate the area of Moorhead, MN, affected by the 100-year flood event (Table
4.13). This includes the time to return home, prepare for the evacuation, travel outside of the affected
area, and reach their destination. The results from the Minnesota side evacuation are summarized in
Table 4.13.

61
Table 4.13 Minnesota Side Evacuation Results
Measure Value
Simulation Planning Horizon (clearance time) 180 Minutes
Number of Vehicles that Reached Destination 16,636 Vehicles
Average Trip Time 14.9 Minutes
Average Trip Distance 8.0 Miles
Average Travel Speed 32.4 MPH

4.3.3.2 North Dakota side evacuation scenario: The input data for the model was prepared in a similar
way to what was done for the Minnesota evacuation scenario. The evacuation population was represented
by people living within the 100-year flood plain on the North Dakota side of the Red River. The hybrid
model estimated that the North Dakota evacuation would take 300 minutes (5 hours), as shown in Table
4.14. The higher evacuation time for this scenario seems logical since the North Dakota side of the Red
River experiences more flooding during the 100-year flood event.

Table 4.14 North Dakota Side Evacuation Results

Measure Value
Simulation Planning Horizon (clearance time) 300 Minutes
Number of Vehicles that Reached Destination 27,459 Vehicles
Average Trip Time 31.2 Minutes
Average Trip Distance 10.3Miles
Average Travel Speed 22.8 MPH

4.3.3.3 North Dakota and Minnesota side evacuation scenario: For this scenario, the population
consists of people living on both sides of the Red River within the 100-year flood plain. The input data
were prepared using the equation developed in a similar way to the previous two scenarios. The
evacuation time estimate for the combined evacuation was 360 minutes (6 hours), as shown in Table 4.15.

Table 4.15 North Dakota and Minnesota Side Evacuation Results

Measure Value
Simulation Planning Horizon (clearance time) 360 Minutes
Number of Vehicles that Reached Destination 35,725 Vehicles
Average Trip Time 34.5 Minutes
Average Trip Distance 11.0 Miles
Average Travel Speed 21.4 MPH

To compare the traffic movement patterns for the different modeled scenarios, the traffic volumes at
several key locations were recorded for each scenario. The results for the previous three evacuation
scenarios are summarized in Table 4.16.

62
Table 4.16 Evacuating Traffic Volumes at Key Locations for the Different Scenarios
Traffic Volumes
Location
Location ND & MN
ID ND Evacuation MN Evacuation
Evacuation
I-94 Red River Crossing EB 1 3,758 321 4,215
I-94 Red River Crossing WB 2 429 3,259 3,777
NP Ave Bridge Crossing EB 3 1,086 150 811
NP Ave Bridge Crossing WB 4 58 836 698
Main Ave Bridge Crossing EB 5 842 145 549
Main Ave Bridge Crossing WB 6 110 836 877
I-94 West of MN 336 EB 7 4,076 1,672 5,339
I-94 West of MN 336 WB 8 8 9 8
I-94 East of Sheyenne St EB 9 238 82 251
I-94 East of Sheyenne St WB 10 1,204 1,122 2,216
I-29 North of 12th Ave N. NB 11 1,851 773 2,546
I-29 North of 12th Ave N. SB 12 1,181 127 1,510
I-29 North of 32nd Ave S. NB 13 437 2 540
I-29 North of 32nd Ave S. SB 14 1,031 687 1,638

The I-94 Red River Bridge crossing served a large number of evacuation trips for all three scenarios. The
highest evacuation volume was observed on I-94 eastbound near MN 336, which is on the east side of the
Fargo-Moorhead metro area. Based on the household evacuation survey, the external zone to the east of
Moorhead had the highest number of responses for a flood evacuation.

4.3.3.4 Evacuation scenario with modified traffic control: To evaluate the effect of traffic operations
measures on the performance of the transportation system during emergency events, an evacuation
scenario was modeled with modified traffic control data. During major flooding event in recent years,
traffic control was performed by personnel from local and state police agencies, National Guard soldiers
and airmen, and volunteers of the Cass County Emergency Vehicle Assistance Communication (EVAC)
organization. In addition to supporting the movement of emergency and essential service vehicles, traffic
control personnel would favor vehicles traveling along the major movements of critical corridors. Local
and state traffic engineers could also develop and deploy traffic signal timing plans to assist movement of
vehicles during evacuation emergencies. To replicate modified traffic control practices, the green time for
the major intersections along several corridors was doubled to facilitate the movement of the evacuation
population (Figure 4.7). The results from this evacuation scenario are summarized in Table 4.17.

63
Figure 4.7 Location of modified traffic signal timing plans

Table 4.17 North Dakota and Minnesota Side Evacuation Results (Modified Control)
Measure Value
Simulation Planning Horizon (clearance time) 330 Minutes
Number of Vehicles that Reached Destination 35,017 Vehicle
Average Trip Time 31.141 Minute
Average Trip Distance 10.629 Mile
Average Travel Speed 22.807 MPH

The output from the model shows that modifying the traffic control during regional evacuation events
helps improve the transportation system performance. By comparing the results from Tables 4.15 and
4.17, there was a reduction of more than 8% in the clearance time needed during the evacuation. In
addition, the reduction in the average evacuation trip time was almost 10%. These tables illustrate how

64
the developed methodology could be used for testing the effects of different traffic management plans to
optimize the use of the transportation system during emergency evacuations.

4.3.3.5 Evacuation scenario using national averages: To evaluate the effect of using data collected for
small- and medium-size MPOs compared with using data developed for large cities for developing
evacuation models, an evacuation scenario was modeled using national data. The new scenario was a
modified version of the flooding on both sides of the Red River, where national trip generation and
loading rates discussed in Section 2 were used.

In this scenario, the number of evacuation trips generated per household was 1.3 vehicles. As for how the
evacuation trips were distributed, the same trip distribution was used since it is based on local area
properties. In addition, the general function developed by Tweedie et al. discussed in Section 2 was used
to load evacuation trips onto the transportation system. The results from this evacuation scenario are
summarized in Table 4.18.

Table 4.18 North Dakota and Minnesota Side Evacuation Results (National Rates)
Measure Value
Simulation Planning Horizon (clearance time) 240 Minutes
Number of Vehicles that Reached Destination 32,687 Vehicles
Average Trip Time 34.0 Minutes
Average Trip Distance 9.6 Miles
Average Travel Speed 17.8 MPH

The results from the running the evacuation scenario show a significant reduction (33.3%) in the
clearance time needed for evacuation compared with the scenario using the data developed for this study.
The main reason for the reduction in the needed clearance time is that evacuation trips are loaded at a
higher rate using the functions available in the literature compared with the function developed in this
study. The functions were illustrated in Figure 3.7. Also, by comparing the results from Tables 4.15 and
4.18, it was shown that using the average trip generation rates underestimated the number of evacuation
trips generated by 8.5%. Hence, the results from this analysis suggest that using human behavior data
from hurricane regions for evacuation modeling might result in underestimating the clearance time and
demand for travel during medium-size metropolitan areas evacuations.

65
66
5. CONCLUSION AND SUMMARY OF THE RESULTS
This section provides a summary of the approach utilized for the purposes of this study. In addition, it
provides the conclusions based on the results of this study. It also presents recommendations for possible
implementation of the modeling approach as well as suggestions for future research.

5.1 Summary of Model Development

Evacuations in cases of regional emergencies result in drastic shifts in the demand for travel over the
transportation system; hence, emergency officials usually report traffic problems during regional
emergency events. Estimating the conditions of the transportation system during regional emergency
events is critical for emergency preparedness, where the development of evacuation plans should be done
well in advance of the occurrence of the emergency event. So far, few studies have addressed evacuation
modeling for small-and-medium size MPOs due to the lack of data and resources.

This study aimed at developing a methodological approach for supporting effective decision making and
testing different emergency scenarios while taking into account the various factors affecting public safety.
The focus of this study was on developing a hybrid evacuation model for urban areas utilizing the
resources available for different metropolitan planning organizations (MPOs) and local data. The
approach developed in this study utilized the widely available travel demand tools at the MPO level as a
modeling basis. These tools were enhanced using simulation class models to provide the required level of
details and allow for testing operational level scenarios.

In this model, the traffic simulation side was integrated with the traditional transportation planning side.
DYNASMART-P software was incorporated with Cube software using an interface that was developed
for this study to introduce traffic dynamics to a traditional planning model. This provided the level of
detail needed to analyze the performance of the transportation system under emergency evacuation
conditions and develop traffic management plans. To illustrate the capabilities of the hybrid model, a case
study was developed using the Fargo-Moorhead metropolitan area regional travel demand model
enhanced by traffic simulation tools.

This study also aimed at developing human behavior data needed for evacuation modeling in medium-
size MPOs and other areas similar in size. The local human behavior data needed for evacuation modeling
were not available in the transportation census or in the literature. That data were needed for trip
generation, trip distribution, and trip loading rates, which are critical for developing accurate and reliable
evacuation models. In addition, the data obtained were used to identify the population compliance rates
with evacuation orders, and the factors affecting the evacuation trip making behavior.

For the purposes of this study, a survey was used to generate the human behavior data needed for
developing the evacuation models. The survey revealed that response rates and behavior of the evacuation
population were different than what was reported for hurricane regions. The effect of the difference in
evacuation behavior between medium-size metropolitan areas and larger cities on the model output was
illustrated through the case study.

To illustrate the modeling approach developed for this study, several emergency evacuation scenarios
were modeled using the Fargo-Moorhead metropolitan area as a case study. The scenarios included
modeling a hypothetical emergency evacuation event of the area within the 100-year flood plain triggered
by flooding of the Red River. For each evacuation scenario, the location, threat level, warning times, and
the affected area’s shape and boundaries were analyzed and determined.

67
The modeling results provided estimates of the time needed to safely evacuate the population out of the
affected area, in addition to different transportation system performance measures, such as average speeds
and travel time for evacuation trips. Also, the developed evacuation model was used to measure the
effectiveness of modifying the traffic controls to facilitate the movement of evacuees out of the affected
area. In addition, the model was used to test the effect of using national human behavior data instead of
the data developed for the purposes of this study on the performance of the modeled transportation system
during evacuation. It was shown that using the national average trip generation rates underestimated the
number of evacuation trips generated. The results from this analysis suggested that using human behavior
data from major metropolitan areas for evacuation modeling might result in underestimating the clearance
time and demand for travel during evacuations of medium size metropolitan areas.

5.2 Research Contributions

This study aimed at developing hybrid evacuation models that recognize MPO’s resources and data
availability to address the need for incorporating the transportation component into their transportation
improvement plans. In addition, this study focused on developing the human behavior data needed for
developing realistic evacuation models, and comparing that data with national evacuation data. This study
demonstrated the possibility of using hybrid models for emergency evacuation planning applications. The
modeling approach developed for this study could be applied at different stages of the emergency
response and preparedness: disaster scenario analysis and mitigation preparedness. The output measures
of effectiveness (traffic volumes, average speeds, and ETEs) associated with different traffic management
options for an emergency evacuation scenario were used to identify the effects of a traffic management
plan implemented for that scenario.

The developed hybrid model is practical with reasonable input data requirements, and it has the ability to
assess regional transportation network performance during regional evacuation events. The modeling
approach used for this study provided direct connectivity with the four-step travel demand model, which
is standard for all MPOs, in addition to the required level of detail for developing evacuation traffic
operations plans. This was achieved by integrating the simulation model within the regional travel
demand model, reducing the complexity associated with updating and maintaining the evacuation model.
The hybrid model structure incorporated a traffic generation component into the model, where some of
the previously developed models didn’t have a travel demand component. Also, many of the previous
models did not model the transportation network in detail and primarily had static supply; those issues
were addressed in this modeling approach.

For the purposes of this study, local human behavior data in response to regional emergency events were
developed using household surveys. These data were not available in the transportation census or in the
literature for small- and medium-size metropolitan areas. The data obtained from the survey were used to
develop evacuation trip generation rates, evacuation trip distribution patterns, and evacuation trip loading
rates. The data generated related human behavior during evacuations for both single-family and multi-
family households. In addition, the data obtained were used to identify the population compliance rates
with evacuation orders and the factors affecting the evacuation trip making behavior.

The human behavior data developed for the purposes of this study showed that response rates and
evacuation behavior of populations in medium-sized metropolitan areas were different from those for
larger cities populations. The analysis of the results from the case study showed that using human
behavior data provided in the literature for large cities in developing evacuation models for medium-size
metropolitan areas generated different results. The modeled scenario that was developed using the
national average human behavior characteristics underestimated the number of evacuation trips and
evacuation time estimates compared with the model developed using local data.

68
5.3 Recommendations for Future Research
Traffic congestion caused by regional emergency evacuations can be life threatening. The sudden increase
in demand will result in excessive loads on roads not typically designed to handle those traffic volumes.
The management of the transportation system during regional emergency events is critical to achieve a
safe and efficient evacuation. The hybrid model developed for this study could be applied at different
stages of the emergency event used to develop traffic management plans to support emergency
evacuations. The study could be expanded in the future for the following:
1) With the capabilities of the new hybrid model, it could be used to test other emergency scenarios,
such as evacuations triggered by hazmat spills and other major incidents. The methodological
approach developed for this study could be applied for different scenarios; however, the trip
generation data need to be modified for each modeled scenario. The structure of the hybrid model
developed for this study makes it efficient to update the travel demand data needed for different
scenarios.
2) There is a great need to generate human behavior data during emergency events for small- and
medium-size metropolitan areas. The results from this study showed that human behavior data
generated for hurricane regions cannot be used to develop accurate evacuation models for medium
size metropolitan areas. The human behavior data could be generated for other metropolitan areas and
the results could be compared with the findings from this study.
3) The developed model didn’t include factors such as age and income levels, which could affect the
evacuation trip making behavior. Further research is needed to capture more factors that affect the
trip making behavior for evacuees.
4) This study didn’t focus on special facilities and transient populations due to their limited effect on the
overall transportation system performance. Separate models could be developed for those population
categories in the future.
5) This study utilized the User Equilibrium principle to assign traffic to their trip routes. This approach
assumes that users have perfect knowledge about traffic conditions on the transportation system and
use the route that minimizes their travel cost. During regional evacuations, the evacuees might use the
road that is most familiar to them, and the model is not capable of capturing social preferences in
traffic assignment. To achieve that, more data need to be collected about the factors affecting the
evacuee’s choice of trip route.

69
70
REFERENCES
1. Advanced Traffic Analysis Center, Upper Great Plains Transportation Institute, “F-M COG 2005
Model Construction & Calibration Technical Document,” Fargo ND, June 2008.
2. Antoine G. Hobeika, and Changkyum Kim, “Comparison of Traffic Assignments in Evacuation
Modeling,” IEEE Transactions on Engineering Management, Volume 45 NO. 2, May 1998.
3. Athanasios K. Ziliaskolpoulos, and Srinivas Peeta, “Foundations of Dynamic Traffic Assignment:
The Past, The Present and the Future,” (Transportation Research Board 81st annual meeting,
Washington, D.C, Paper #02-2861, January 2002).
4. Becky P. Y. Loo, “Role of Stated Preference Methods in Planning for Sustainable Urban
Transportation: State of Practice and Future Prospects,” Journal of Urban Planning and
Development, p.p. 210-224, December 2002.
5. Bing Mie, “Development of Trip Generation Models of Hurricane Evacuations,” (Master’s Thesis at
Louisiana State University, Department of Civil and Environmental Engineering, Louisiana, August
2002).
6. Brian Wolshon, “Empirical Characterization of Mass Evacuation Traffic Flow,” (Transportation
Research Board 87th annual meeting, Washington D.C., January 2008).
7. Bridget Barrett, Bin Ran, and Rakha Pillai, “Developing a Dynamic Traffic Management Modeling
Framework for Hurricane Evacuation,” (Transportation Research Board 79th annual meeting,
Washington D.C., January 2000).
8. Cambridge Systematics, Travel Survey Manual, Prepared for U.S. Department of Transportation,
U.S. Environmental Protection Agency, July 1997.
9. FHWA Web site, “Catastrophic Hurricane Evacuation Plan Evaluation: a Report to Congress,” U.S.
DOT in Cooperation with U.S. Department of Homeland Security, June 1, 2006.
10. Citilabs Web site, http://www.citilabs.com, accessed on December, 2008.
11. CSEPP Accident Planning Base Review Group, “Emergency Concept Plan for the Chemical
Stockpile Emergency Preparedness Program,” 1997.
12. Daniel L. Dornan, and M. Patricia Maier, “NCHRP report 525: Surface Transportation Security,
Incorporating Security into the Transportation Planning Process,” Volume 3, Transportation Research
Board of National Academies, 2005.
13. David A. Hensher, and Kenneth John Button, Handbook of Transport Modeling, Volume 1, (Emerald
Group Publishing, 2000).
14. Edmond Chang, “Traffic Simulation for Effective Emergency Evacuation,” (International Chinese
Transportation Professionals Association, 4th Asia-Pacific Transportation Development Conference,
Oakland, CA, April 2003).
15. John C. Whitehead, “Environmental Risk and Averting Behavior: Predictive Validity of Jointly
Estimated Revealed and Stated Behavior Data,” Appalachian State University, Boone, NC, Article
provided by European Association of Environmental and Resource Economists in its journal
Environmental & Resource Economics, Volume 32, 2005, Issue 3, November, p.p. 301-316.
16. Fargo-Moorhead Metropolitan Council of Governments growth projections for the Fargo-Moorhead
Metropolitan Statistical Area, 2005.
17. Federal Highway Administration, Model Validation and Reasonable Checking Manual, Washington
D.C., February 1997.
18. Federal Highway Administration, Traffic Analysis Toolbox Volume III: Guidelines for Applying
Traffic Microsimulation Modeling Software, FHWA-HRT-04-040, July 2004.
19. Fink, A., The Survey Kit: How to Sample in Surveys, (Sage Publications, Thousand Oaks, CA, 1995).

71
20. Hani S. Mahmassani, Hayssam Sbayti, and Xuesong Zhou, DYNASMART-P, Intelligent
Transportation Network Planning Tool, Version 1.0 User’s Guide, Maryland Transportation
Initiative, University of Maryland, College Park, September 2004.
21. Institute of Transportation Engineers, Trip Generation Manual, Seventh Edition, Washington D.C.
2003.
22. Irwin, M., and J.S. Hurlbert, “A Behavioral Analysis of Hurricane Preparedness and Evacuation in
Southwestern Louisiana,” Louisiana Population Data Center, September 1995.
23. John Sorensen and Barbra Vogt, Interactive Emergency Evacuation Guidebook, prepared for the
protective action IPT chemical stockpile emergency preparedness program, February 2006.
24. Lianyu Chu, Henry X. Liu, Jun-Seok Oh, and Will Recker, “A Calibration Procedure for Microscopic
Traffic Simulation,” (Transportation Research Board 83rd annual meeting, Washington D.C., January
2004).
25. Louviere, Jordan J., Combining Revealed and SP Data: The Rescaling Revolution, Sydney, Australia:
The University of Syndey, NSW 2006.
26. Mc Clave, Sincich, Statistics, Tenth Edition, Pearson Prentice Hall, New Jersey 2006.
27. Michael G. McNally, “The Four Step Model,” The University of California Transportation Center,
Berkeley, Paper UCI-ITS-AS-WP-00-5, 1994.
28. Michael K. Lindell, and Carla S. Prater, “Critical Behavioral Assumptions in Evacuation Time
Estimate Analysis for Private Vehicles: Examples from Hurricane Research and Planning,” Journal of
Urban Planning and Development, March 2007.
29. Michel D. Meyer, and Eric J. Miller, Urban Transportation Planning a Decision Oriented Approach,
2nd edition, McGraw-Hill, New York, New York, 2001.
30. Municipal Airport Authority of the City of Fargo, Small Community Air Service Development Grant
Application, Fargo, ND, April 2005.
31. National Association of Regional Councils, “Complying with SAFETEA-LU: Thirteen Items MPOs
Must Address,” Washington, D.C., June 1, 2006.
32. National Cooperative Highway Research Program, Report 365, Transportation Research Board,
“Travel Estimation Techniques for Urban Planning,” Washington, D.C. 1998.
33. National Cooperative Highway Research Program, Report 392, Transportation Research Board,
“Transportation’s Role in Emergency Evacuation and Reentry,” Washington, D.C. 2009.
34. Nicholas B. Taylor, “The CONTRAM dynamic traffic assignment model,” Networks and Spatial
Economics 3, p.p.297-322, 2003
35. Pamela M. Murray, and Hani S. Mahmassani, “Model of Household Trip Chain Sequencing in an
Emergency Evacuation,” (Transportation Research Board 82nd annual meeting, Washington D.C.,
January 2003).
36. Post, Buckley, Schuh & Jernigan, Inc. (PBS&J), “Southeast Louisiana Hurricane Evacuation Study:
Abbreviated Transportation Model Development and Format,” draft, Tallahassee, FL, April 2001.
37. Post, Buckley, Schuh & Jernigan, Inc. (PBS&J), “Southeast United States Hurricane Evacuation
Traffic Study: Behavioral Analysis,” Technical Memorandum 1, Final Report, Tallahassee, FL, May
2000.
38. Ramachandran Balakrishna, Constantinos Antoniou, and Haris N. Koutsopoulos, “Calibrating Speed-
Density Functions for Meso-scopic Traffic Simulation,” (Transportation Research Board, Traffic
Flow Theory and Characteristics Committee (AHB45), 2008 Meeting, Woods Hole, MA, 2008).
39. Ramachandran Balakrishna, Haris N. Koutsopoulos, Moshe Ben-Akiva, Bruno M. Fernandez Ruiz,
and Manish Mehta, “A Simulation-Based Evaluation of Advanced Traveler Information System,”
(Transportation Research Board 84th annual meeting, Washington, D.C, Paper #05-2316, 2005).

72
40. Reuben B. Goldblatt, and Kevin Weinisch, “Evacuation Planning, Human Factors, and Traffic
Engineering,” Transportation Research News, Issue 238, p.p. 13-17, May-June 2005.
41. Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (SAFETEA-
LU), Public Law No109-59, Washington, D.C. 2005.
42. Sirinivasan Sundaram, “Development of a Dynamic Traffic Assignment System For Short-Term
Planning Applications,” (Master’s Thesis at the Massachusetts Institute of Technology, Department
of Civil and Environmental Engineering, MA, June 2002).
43. Sirnivas Peeta, and Hani S. Mahmassani, “Multiple User Class Real-Time Traffic Assignment for
On-Line Operations: A Rolling Horizon Solution Framework,” Transportation Research Part C:
Emerging Technologies, Volume 3, Issue 2, pp. 83-98, 1995.
44. Southworth, F., “Regional Evacuation Modeling: A State-of-the-Art Review, Center for
Transportation Analysis,” Oak Ridge National Laboratory, Oak Ridge, TN, March 1991.
45. The Advanced Traffic Analysis Center Web site, www.atacenter.org, accessed on February 2009.
46. The City of Fargo Web site, http://www.cityoffargo.com/, accessed on January 2009.
47. Transportation Research Board, Highway Capacity Manual 2000, Washington D.C., 2000.
48. TRB Subcommittee ANB10(3) Emergency Evacuation Web site, http://www.rsip.lsu.edu/anb10-
3/Resources/resources.htm , accessed February 1, 2009.
49. United States Army Corps of Engineers, Federal Emergency Management Agency, National Oceanic
and Atmospheric Administration, National Hurricane Center, Mississippi Emergency Management
Agency, Mississippi Hurricane Evacuation Study Technical Data Report, December 2001.
50. USDOT, “Calibrating and Testing a Gravity Model for any Size Urban Area,” August 1983.
51. Wilco Burghout, “Hybrid Microscopic-Mesoscopic Traffic Simulation,” (Doctoral Dissertation at
Royal Institute of Technology, Stockholm, Sweden, 2004).
52. William Mendenhall, and Terry Sincich, “A Second Course in Statistics: Regression Analysis,” 6th
Edition, Pearson Education, Upper Saddle River, NJ, 2003.
53. Yi-Chang Chiu, Pavan Korada, and Pitu B. Mirchandani, “Dynamic Traffic Management for
Evacuation,” (Transportation Research Board 84th annual meeting, Washington D.C., January 2005).
54. 23 Code of Federal Regulations (CFR), Title 23: Highways, Part 450.322, April 1, 2003.

73
74
APPENDIX A: Fargo-Moorhead Emergency Evacuation
Survey

75
 NDSU RESEARCH STUDY
Emergency Evacuation Survey

Dear resident:

My name is Shawn Birst. I work for the Upper Great Plains Transportation Institute at North Dakota State
University, and I am conducting a research project to collect data on residents’ response and travel
behavior related to a hypothetical emergency evacuation event triggered by river flooding. The results of
this study will be used to assist metropolitan emergency management and transportation agencies to
improve preparedness in cases of emergencies.

You are invited to participate in this research project. Your participation is voluntary, and you may
decline or withdraw from participation at anytime without penalty. If you have any questions about this
project please contact me at shawn.birst@ndsu.edu. If you have any questions about the rights of human
participants in research or to report a problem, contact the NDSU IRB Office, (701) 231-8980, or
ndsu.irb@ndsu.edu.

A-1
 Emergency Evacuation Survey
Introduction: The Upper Great Plains Transportation Institute (UGPTI) at NDSU in partnership
with Cass and Clay County Emergency Management centers is conducting this survey to collect
data on residents’ response and travel behavior related to a hypothetical emergency evacuation
event triggered by Red River flooding. The data will be used to assist metropolitan emergency
management and transportation agencies to improve preparedness in cases of emergencies. Your
feedback will help improve the safety of your community. It should be noted that all the
information provided will be kept confidential. Please mail back your completed survey using
the provided return envelope by November 1, 2008.

Evacuation Scenario: Immanent (within a few hours) Red River flooding is forecasted
(e.g., levee breach and/or overflow) resulting in significant flooding for your area that
would not recede for several days. Therefore, evacuation orders are issued by the local
jurisdictions. Based on this scenario please complete the following survey.

Please respond to the following eight questions in a manner that best applies to your case:

Q1) If mandatory evacuation orders are issued by the authorities, would you:

Comply with evacuation orders

Not comply with evacuation orders. Please list the reason(s) for not complying:

_________________________________________________________________

Q2) When the evacuation orders are issued, how much time do you need to prepare before
leaving the household?

(1-15) minutes

(16-30) minutes

(31-45) minutes

(46-60) minutes

(>60) minutes

A-2
Q3) How many vehicles would you use during the evacuation process?

None

One

Two

Three or more

Q4) If you answered “None” to Q3, how would you be evacuated?

Relative or friend

Local authority/agency

Q5) When the evacuation orders are issued, where would you evacuate to?

Relative or a friend’s house

Hotel or motel

Public shelter

Q6) Referring to the attached map, please write the zone number that you would most likely
evacuate to?

Zone _____

Q7) How would you classify your residency?

House (single family dwelling unit)

Apartment (multi-family dwelling unit)

Q8) You prefer to be notified about the emergency by:

Television alert system

Radio alert system

Telephone alert system

Other

Upper Great Plains Transportation Institute • North Dakota State University

North Dakota State University does not discriminate on the basis of race, color, national origin, religion, sex,
disability, age, Vietnam Era Veterans status, sexual orientation, marital status or public assistance status. Direct
inquiries to the Executive Director and Chief Diversity Officer, 202 Old Main, (701) 231-7708.

A-3
A-4
80
APPENDIX B: Fargo-Moorhead Emergency Evacuation
Survey Results

81
Evacuation Modeling for Small to Medium Sized Metropolitan Areas
Table 1. Households Compliance with Evacuation Orders
# Single-Family Comply 389 # Single-Family Not Comply 25
# Multi-Family Comply 23 # Multi-Family Not Comply 0
# Total Households Comply 412 # Total Households Will Not comply 25

Table 2. Compliance Rates by Household Category

Household Type % Comply % Not Comply Total
Single-Family HH 94.0% 6.0% 100.0%
Multi-Family HH 100.0% 0.0% 100.0%

Table 3. Time Needed to Evacuate

1-15 min. 16-30 min. 31-45 min. 46-60 min >60 min.
23 88 80 112 109
5.6% 21.4% 19.4% 27.2% 26.5%

Table 4. Number of Vehicles Used for Evacuation by Household Category

# of Vehicles # of Single-Family % of Single-Family # of Multi-Family % of Multi-Family
Used HH HH HH HH
0 0 0.0% 0 0.0%
1 171 44.0% 18 78.3%
2 187 48.0% 5 21.7%
>=3 31 8.0% 0 0.0%
Total 389 100.00% 23 100.0%
Average/HH 1.64 Vehicles/HH 1.22 Vehicles/HH
Average vehicles used = 1.62

Table 5. Evacuation Destination

Destination Choice # of Households % of Households
Relative or Friend's House 318 77%
Hotel or Motel 70 17%
Public Shelter 24 6%
Total 412 100%

B-1
 Table 6. Evacuation Destination by Household Category.
Zone 1 2+3 4 5 6+9
Single-Family HH 9.3% 1.3% 6.4% 6.4% 2.8%
Multi-Family HH 4.4% 4.4% 8.7% 0.0% 4.4%
Zone 7 8 10 11+13+15 12
Single-Family HH 3.6% 3.1% 1.5% 1.3% 2.6%
Multi-Family HH 4.4% 0.0% 4.4% 4.4% 4.4%
Zone 14 16 17 18 19
Single-Family HH 4.1% 12.1% 12.1% 6.9% 26.5%
Multi-Family HH 4.4% 21.7% 4.4% 8.7% 21.7%

Figure 1. Red River Flooding Evacuation Zones

Table 7. Preferred Method of Notification for Emergency Evacuation.
Destination Choice # of Households % of Households
Television alert system 210 49.6%
Telephone alert system 162 38.3%
Radio alert system 51 12.1%
Total 423 100.0%

B-2

View publication stats